The server-side sync and the pull-request checks used to be GitHub only; on GitLab the answer was the CLI in your pipeline. GitLab now connects the same way GitHub does.
In the web app: Settings → Repositories has a GitLab card next to GitHub — connect once, then connect its repositories to your projects as you would a GitHub repository
On GitLab: connecting a project opens a merge request that adds it to meldoc.config.yml. Once that is merged, the documentation syncs from the default branch, and an edit made in Meldoc comes back as a merge request
On GitLab: merge requests get the Meldoc checks as a commit status, with the report as a comment
In the web app: if Meldoc's access to GitLab is revoked, the project says Reconnect needed instead of looking synced, and reconnecting restores everything without touching the project
Bitbucket: still the CLI in your pipeline for now
Weekly Reports
By email: on Monday morning, one report per workspace lists the projects you watch that changed last week — how many documents changed, were added or removed, how many people changed them, and links to the most-changed ones. Nothing changed, no email
In the web app: you watch the projects you maintain automatically, and anyone can watch a project, or stop, with the eye button next to Share on its overview
In your account: Weekly reports under Email notifications turns the reports on and off, and every report has an unsubscribe link
Published Documentation
On your own site: on Enterprise, a project's documentation can open at an address on your site, such as https://example.com/docs. Enter it under Project settings → Canonical URL, point a reverse proxy you control at the address shown under Point your proxy at, and select Check. Links, images and search stay under your address, and once it reads Verified, the pages on public.meldoc.io tell search engines your address is the original
In search engines: public projects now have a sitemap, so search engines find their documents. Documents behind a password and archived ones are left out
In link previews: a shared link to a public document shows its first paragraph as the description — never for a document behind a password
Repository Sync
In the web app: the repository panel says how many documents changed in Meldoc since the last export and when the next export runs, and Push to repository sends them now
In the web app: a document that changed both in Meldoc and in the repository, or was removed in the repository while it had edits here, is marked in the tree and says above its content why it does not sync. Project admins get a Resolve the conflict link
In the web app: deleting a project disconnects its repositories at once, so it no longer gets in the way of other projects on the same repository
Workspaces and Teams
In the web app: any member can leave a workspace under Settings → Danger zone → Leave workspace. The owner cannot leave or be removed, and the last admin has to make someone else an admin first
In the web app: anyone in a team can leave it with Leave team in the team's dialog; only an admin can add them back
In the web app: deleting or leaving a workspace takes you home instead of to a "not found" page
Smaller Changes
On GitHub and GitLab: after Mark as wrong, a button takes you back to the pull request or merge request you came from
AI: AI features moved to newer models. If your workspace uses its own AI provider key, make sure it has access to the provider's latest models
Billing: invoices and receipts go to the billing email, which the owner changes under Settings → Plan & Billing → Open portal. The owner is now also emailed while a payment is overdue
Documentation Review
The review that reads a pull request against your documentation now says exactly what it checked, remembers what it said last time, and takes no for an answer.
On GitHub: every outdated sentence in the review's comment carries a Mark as wrong link. It opens Meldoc with the quoted sentence, and once you confirm, the review stops reporting that sentence for the repository until somebody edits it. Marking needs permission to edit the project's documents
On GitHub: each push still reviews the whole pull request, but remembers the last run — a finding still there is marked still open since the commit it first appeared on, one that went away is struck through under Resolved since the last run with the reason, and one it could not check again is listed as not re-checked
On GitHub: a run that could not check every document no longer says "Docs are up to date". It concludes neutral, its title reads Some docs not checked with the count, and its comment opens with Not every doc was checked
On GitHub: findings on one changed line share one annotation, placed on the line that changed rather than the top of the hunk, and a line another project of the same repository already annotated gets no second one
On GitHub: fixing the document in the same pull request counts — the review reads your edited version
On GitHub: changed tests are left out of the review, and a sentence the pull request wrote itself is never reported as outdated by it
On GitHub: the review and the pre-merge check run only on pull requests into the default branch, the one Meldoc syncs — a release pull request into main when develop is the default gets neither, so mark the checks required on the default branch only
In the web app: since the end of the month a project chooses which pull requests get both checks — Check pull requests into in the sync panel takes the default branch only, every branch, or the default branch plus patterns such as develop or release/*
The two GitHub checks are renamed: Meldoc: <project> is now Meldoc sync: <project>, and Meldoc docs: <project> is now Meldoc review: <project>. If you marked either as required in branch protection, change the rule to the new name
After the merge: findings no longer stay behind in the pull request. Open ones move onto the documents in Meldoc, above the title, and the sentence is highlighted — Code changed, doc didn't when the pull request was merged into the synced branch, Changes after release when it went into another branch such as develop, Fix planned once someone selects Fix later, and a quiet Checked against code when a merged pull request was reviewed and nothing was found. A finding closes when the sentence is edited, or when you select All correct
In the web app: in a document's list of findings, the file path links to that line in the pull request's diff, and the pull request number to the pull request
On GitHub: a push that changed nothing the review reads spends no AI requests and says so, and the same finding no longer comes and goes between pushes of unchanged code
On GitHub: a finding the review could not verify is labelled Not double-checked instead of passing for a confirmed one
Search with AI Answers
Search used to hand you a list and leave the reading to you. Now it answers the question from the documents it found, and shows which ones.
In the web app: an AI answer appears above the results, written only from the documents in them — the same ones you see, with the same filters. They are listed under it as Sources, and a marker like [1] links to the source it came from
Shows nothing rather than guess: there is no answer to a lookup — an alias, a project::doc reference, an id or a path — or when the results do not contain one
In the web app: with your workspace's own AI provider, the answer appears by itself once the results settle; on Meldoc's key it waits for Get an AI answer or Enter and counts against your plan's AI requests. The AI answer switch turns answers off altogether
In the web app: a project search that finds nothing, or only weak matches, looks through your other projects too and lists what it found under Found in other projects — kept apart, so the filter you chose still holds, and never used for the answer
In the web app: the workspace overview and every project overview have a search field at the top, with results dropping down under it — filtered to the project on a project's overview
In your AI assistant: docs_search returns those matches from other projects in a separate elsewhere list, so an assistant can point you to the right project without mixing it into this one's results
AI Usage
In the web app: Settings → AI Provider shows what your own key spent, one month at a time — requests, tokens in and out and an approximate cost at list prices, per feature and model, for the last 13 months. Only workspace admins see it, and your provider's console still has the bill
Start Page and Project Health
Opening a workspace now answers "what needs me?" before you go looking.
In the web app: the workspace overview opens on a Start tab. Needs you lists the comment threads waiting for your reply and the comments that mention you, and, for the projects you maintain, documents the code check flagged, broken links, stale aliases and repositories that need attention. Recently updated shows the last two weeks of changes — a sync from the repository as one line — next to the documents you viewed recently
In the web app: every project overview has a Project health block — repositories, code check, broken links, stale aliases and open comments, one line each, with a dot when something is wrong and a quiet check when all is well. Each line opens in place
In the web app: open comment threads are visible from the project, grouped by document with the sentence they are about, and one click opens the thread in the document
In the web app: Code check and Repositories are tabs of the project overview. Code check opens with one sentence — how many documents say something the code no longer does — and groups the findings by document
Comments
In the web app: type @ in a comment to mention someone in the workspace. A list of people opens under the field, the mention shows as a chip, and the person sees it under Needs you on their Start page until they reply
In the web app: links in comments are clickable and open in a new tab
In your AI assistant: an assistant can mention people too, finding them with workspace_members_list, and commenting over MCP needs read access to the document — the same as in the app
MCP on Every Plan
In your AI assistant: creating and editing documents over MCP, and a token for the whole workspace, are no longer Pro-only — Free and Team get them too, as the REST API and the CLI already did. Requests per minute still follow the plan
Connecting a Repository
Connecting a repository used to end with "now edit meldoc.config.yml". Meldoc now writes that file for you, as a pull request, and a project's Repositories page shows every connected repository at once.
In the web app: Connect repository asks which repository, which folder holds the project's documentation, and which checks to post — the pre-merge check starts switched on and the documentation review off, and either can change before you confirm
On GitHub: Meldoc opens a pull request on the branch meldoc/config that adds the project to meldoc.config.yml — creating the file, adding to its list, or moving a single-project config to the multi-project form. It never commits to your default branch, keeps the file's comments and key order, and a project connected while it is open adds a commit instead of a second pull request
In the web app: when the file cannot be changed safely — the folder overlaps another project's, or the existing project sits at the repository root — the dialog says why, and you can still connect and edit the config yourself
In the web app: each connected repository has its own card with the sync state, the config pull request, both checks and any conflicts, so nothing about a pairing is behind a tab; until the config is merged, the card says Waiting for the config pull request and links to it
In the web app: a workspace owner installs the GitHub App from a Connect a repository card on the dashboard or from a project's empty Repositories page, and can pick a repository the workspace has not allowed yet straight from the connect dialog
In the web app: the workspace's repository list is two sections, Allowed for projects and On GitHub, not allowed yet
Getting Started
A new workspace used to open on an empty page. Signing up now walks you through setup, and the first project explains Meldoc in three pages.
In the web app: setup asks for your name and the Terms of Service, then the workspace — created on Free, with the upgrade offered on the next step — then teammates to invite on a paid plan, the GitHub App, an AI assistant and your email preferences. The invites, the GitHub App and the assistant can be skipped, and both kinds of email start off
In the web app: creating another workspace takes the same steps, and ✕ leaves at any of them — back where you were before the workspace exists, into the new workspace on Free after
In the web app: the Getting Started (demo) project is now three pages — Welcome, Connect your repository, Connect your AI assistant — and does not count toward your plan's project or document limits; documents you add to it do
In the web app: Create project carries on past the name — connect a repository folder, pick a starting set of document types (Engineering, Product, Knowledge base, or Blank) and add contributors. Every step after the first can be skipped, and ✕ keeps what you have done so far
Document Types and Fields
The app now says document type everywhere it used to say template, because that is what one is: a type decides which fields a document has, never its text. The template: key in front matter is unchanged.
In the web app: a field can carry a description of up to 280 characters on what to write in it — an info icon next to the label shows it on every document, and AI assistants read it before they fill the field in
In the web app: which documents a field applies to is set in the field's own dialog — All documents or Only these document types — and the order and layout of the fields documents share live on the Custom fields tab, instead of in a built-in "Default" type
In the web app: changing a document's type asks first when fields with values would stop applying, and so does narrowing a field; the values are kept either way and come back if the field applies again
In published documentation: a document's custom fields appear under its title, in the same order and layout as in the app, and the tree shows each type's icon
Published Documentation
Documentation published at public.meldoc.io now looks like meldoc.io, which was redesigned this month too — and reads exactly as before: the same text size, line height and column width.
For your readers: a link to any page unfurls in chat apps and social posts with its title and a "workspace › project" line, and no Meldoc image on your documents
For your readers: the theme follows the system unless they pick one with the single theme icon
For your readers: the start page offers Continue reading from pages read in that browser — kept in the browser only, and never including a password-protected document
Fixed: links between documents render again, images load, and archived documents stay out of published lists
Git Sync
A two-way sync has to stay two-way. Most of this month's sync work is about that.
In the web app: a document the sync wrote is no longer locked for editing — every import used to lock it, which quietly made the sync one-way. Documents locked that way earlier stay locked until you unlock them
In the repository: renaming or moving a .meldoc.md file moves the document instead of deleting it and creating a new one
In the web app: a document's History tab tells the sync's side of the story — an edit sent back reads Sent to owner/repo in PR #N, a conflict is recorded when it starts waiting, and an import names the fields it changed and links its commit
In the web app: archiving a document or renaming the alias of a synced project says what will happen in the repository before you confirm
In the web app: a pairing that syncs nothing is no longer shown as Synced, and the conflict count shows only conflicts that are still open
On GitHub: the sync's pull request survives a squash merge of the one before it, and an edit in Meldoc no longer renumbers untouched documents in the export
CLI 1.4.1–1.4.4
Four releases of the meldoc binary, so everything in this section is from the CLI.
meldoc updateVersion installs only what it can verify: a release whose checksum cannot be confirmed is refused, and the installed binary is left untouched
meldoc updateVersion works on office and CI networks: it no longer fails when many people update from one network
Staying signed in: running the MCP server and a command at the same time no longer signs you out, and a lost connection no longer logs you out
meldoc push merges conflicts more reliably: a resolved conflict no longer reads as undoing the other side's edit
File names that work on every system: a pulled document's file name is lowercase and free of characters Windows refuses, and two aliases that differ only in letter case no longer share one file on macOS and Windows
Two agent skills instead of four: meldoc-format for front matter, ordering and links, and meldoc-writing for the text, whether a document is a file or written over MCP. meldoc skills update swaps them in, and each is downloadable at meldoc.io/skills/<name>/SKILL.md
meldoc validate checks ordering: order-sequence warns when a parent's children are not numbered from 0 without gaps
meldoc validate accepts custom fields again: the fields: block no longer reads as an unknown key, so a document with custom fields validates clean
Smaller Changes
In the web app: renaming a project's alias keeps the old one as a redirect inside Meldoc, and the form says why an alias is refused instead of quietly disabling the button
In your AI assistant: docs_create and docs_update take isLocked to lock or unlock a document, and docs_get reports it
In the web app: a page whose data failed to load says so instead of breaking
In the web app: the Relationships graph puts the hub in the middle and leaves archived documents out at every zoom
In the web app: a link to a heading with non-Latin characters scrolls to it
In the web app: paying for an upgrade brings you back to your workspace instead of a "not found" page
In the web app: analytics cookies can be switched off at any time under Privacy in your profile — Accept and Decline carry the same weight, and nothing is counted before you accept
Security: tighter protection for password-protected pages and for changes made through an AI assistant
GitHub Sync
Keeping documentation and a repository in step used to mean a CI job somebody had to write and then keep alive. Now you connect a project to a repository and the two stay in step on their own — nothing ever commits straight to your default branch, and nothing is deleted when you stop.
In the web app: install the GitHub App once for the workspace under Settings → Repositories and choose which repositories it may see, then pair any of them with a project from the project's new Repositories panel — one repository can feed several projects, and each pairing is independent, with its own runs and its own conflicts
In the repository: whatever lands on the default branch arrives in the project — new documents, edits, deletions — picked up on push and again on a schedule, so a lost notification costs minutes rather than a missed change
On GitHub: edits made in Meldoc come back to you as one open pull request on a meldoc/sync/<project> branch, collecting further edits as new commits until you merge it — never a direct commit, never a force-push
In the web app: when both sides changed the same file, the document is marked as a conflict and waits for a person while the rest of the run carries on; open View diff in the panel, then take the repository version or keep Meldoc's — conflict markers never end up in a page
In the web app: every run is listed in the panel, newest first, with what it did; a very large change is worked through across several partial runs, and a partial run never deletes anything
In the repository: several repositories can feed one project — an API, a worker and a mailer each owning a section of one documentation tree — with scope in each repository's meldoc.config.yml saying which part is its business
In the web app: the panel shows what each subscription is limited to, read from the config on every run, so a missing or mistyped scope is visible without opening the repository
In the web app: stop at three sizes — unpair one project from its panel, take a repository out of the pool, or disconnect GitHub entirely under Settings → Repositories — and nothing is deleted on either side
On every plan: repository sync is available everywhere, and a workspace where it is switched off says so instead of hiding the panel
Two Checks on Every Pull Request
A conflict you find after merging is one somebody has to come back to, and a sentence a change made untrue is one nobody comes back to. Two GitHub checks now catch both while the pull request is still open. Both are off by default, switched on per project from the project's Repositories panel, and keep a fixed name so you can make them required in branch protection.
On GitHub: the pre-merge check, named Meldoc: <project>, runs on the pull request and reports what the sync would do with it — a clean merge, a conflict it cannot decide, or front matter that does not parse — as annotations on the file and line they concern, so the author reads them where they are looking
Never writes anything: no document, no conflict, no uploaded image — a pull request that never merges leaves no trace in Meldoc
On GitHub: the summary always prints what it checked and what it could not, because a green check that skipped something is worse than no check
On GitHub: the documentation review, named Meldoc docs: <project>, asks the other question — does this change make something the documentation says untrue? It reads the code the pull request changed against the project's documents and reports dead references and refuted claims, quoting the document's own words verbatim
On GitHub: the review never fails the merge — a sentence a model doubted is not a reason to block a branch, so findings conclude neutral, none concludes success, and the verdict is there to be read
On GitHub: its boundary is part of the report — documents outside the project, code the pull request did not touch, and a diff GitHub sent only in part are all named in the summary, with counts
In the web app: judging meaning needs the workspace's own AI provider, connected under Settings → AI Provider
On GitHub: without a key of your own the review still posts, reports dead references, and says in its summary that it could not judge meaning
AI Provider
Meldoc's AI features can now run on a key you own, and Anthropic joins OpenAI as a provider — so the model reading your documentation is one you chose and pay for directly.
In the web app: connect either provider under Settings → AI Provider and pick a model from the short list Meldoc actually runs — every AI task here asks for structured output, and the dropdown only offers models that hold to it
In the web app: the key is verified with the provider before it is stored, so a typo, a revoked key, or an organisation without access to a model is an error on the form rather than a silent failure in a background run
On GitHub: the documentation review's pull request comment ends with what the run cost — the model, tokens in and out, an approximate cost at list prices, and the total across every review of that pull request
You pay the provider directly: Meldoc neither bills nor meters usage on your own key
Contributors
Every document now shows who wrote it — not only who saved it. The person who ran the push is not always the person who wrote the file, and git already knows the difference.
From the CLI: meldoc push credits whoever git says wrote each file — the author of its last commit plus that commit's Co-authored-by trailers
In the repository: a change merged into a repository Meldoc syncs credits whoever git says wrote each file in that push, the same way
In the web app: those names land in a Contributors row on the document
In the web app: each entry in version history carries its own co-authored by line under the author, so who saved and who wrote are both on record
In the web app: Settings → Members → Git accounts lists every address seen in commits; link one to a member and their name and avatar appear on every document that account ever touched — several addresses collapse into one person
Linked automatically when it is safe: a verified address that matches a member, or a GitHub users.noreply address in its numbered form, is linked without you; everything else waits for an admin
In the web app: open the list from the avatars to hide a bot or a stray name, or credit a workspace member by hand — being listed gives no access, takes no seat, and notifies nobody
Document Templates
A project's custom fields used to apply to every document, whether they made sense on that page or not. A template gives a document a type, and with the type come exactly the fields that kind of page needs, in the order and layout that kind of page reads best.
In the web app: a project defines its templates — a title, a description and an icon, plus the custom fields that belong to that type. Each type arranges its fields in its own order and its own layout: a table, an inline row or a stacked list
Fields that belong to a type: assigning a template is what makes its fields applicable — a runbook stops carrying an API endpoint's fields. Changing a document's type never deletes a value: a field that stops applying comes back with its value if the type is put back
In the web app: a field can hide itself on documents where it is empty, so an optional field stops showing as a blank row
In the repository: a document names its template with template: in its front matter, and meldoc pull keeps that key honest in both directions — written when the server has one, removed when it does not
From the CLI: meldoc templates list and meldoc templates show browse the project's types and the fields each carries, and meldoc new --template writes the key — offline too, from a cache pull refreshes
From the CLI: a typo cannot fail a push — an unknown template alias is a warning, and the document's existing assignment is left as it was
Nothing changes without templates: a project that defines none behaves exactly as before
Import
Documentation that already lives in Notion or in a folder of Markdown can now come across as files you read before anything reaches the server — so a conversion that went wrong is a diff to fix, not pages to delete by hand.
From the CLI: meldoc import --from notion reads an official Markdown & CSV export, and --from markdown reads any directory — a docs-as-code repository, a tree of README.md, an Obsidian vault
In the repository: the import writes local .meldoc.md files in the same layout meldoc pull produces and never touches the server, so you read the diff and run meldoc validate before anything is pushed
In the repository: links between pages become [[alias]], referenced files are copied into assets/, and non-Latin titles are transliterated into readable aliases
In the repository: MIGRATION_REPORT.md lists what the target syntax could not express — Obsidian callouts, Notion databases, files missing from the export — each with its source file
Duplicate Radar
Similar is not the same as redundant: a dozen pages built from one template look alike and should stay that way. Duplicate Radar now says why two documents came out similar, so the clusters worth merging are the ones you see first.
In the web app: a cluster can carry a verdict — same subject, twice, shared block, parallel structure, not really similar, or could not tell — and on the Duplicates tab the ones that say "merge" rise to the top while the "leave it alone" ones sink and arrive collapsed
Certain verdicts clear the cluster: a parallel structure or not really similar cluster Meldoc is confident about disappears from the tab, the document, and what an assistant sees over MCP — and comes back on its own if the pages change; nothing that suggests an action is ever hidden
In the web app: dismiss a match on the document or its cluster on the tab and it is hidden for the whole workspace; Dismissed only finds it again to Restore
In the web app: Recheck looks at a cluster again straight away; otherwise verdicts refresh in the background, at most once a day per cluster
CLI 1.3.4 and 1.4.0
Two releases of the meldoc binary, so everything in this section is from the CLI. The larger change is that meldoc validate now audits the documentation set as a whole, with rules that each exist because that regression actually happened.
meldoc validate audits the set: nine structural rules catch what used to publish silently — two block a publish (a front-matter key Meldoc does not read, and a page that would land at the tree root because nothing declares its parent), the rest advise: duplicate headings, a page rebuilt from a template's container headings, a heading, list or table glued to the paragraph above it, a table row with the wrong column count, and two pages sharing a quarter of their text
validate --fix inserts the blank lines: the spacing warnings are repaired by insertion only — it never removes, reorders or reformats, so it is safe to run in a pipeline
A failed validation names the rule: Validation failed with 2 error(s): unknown-field, orphan instead of one fixed sentence about aliases, so you know what to fix before you open a file
meldoc pull keeps your key order: front matter is no longer rewritten alphabetically — that had read as a Meldoc-side edit and turned a repository change to the same lines into a conflict
meldoc push credits the authors: the git author and co-authors of each file travel with the push — and a push from a plain directory with no git still publishes, crediting nobody
Archive
Some pages are no longer true but still worth keeping. Deleting them breaks every link, leaving them misleads readers — archiving is the third answer.
Retire without deleting: an archived document keeps its content and every link to it keeps resolving — it just leaves the sidebar tree, search, and assistant listings
Read-only, on purpose: content, name, alias, and fields freeze until you restore it, so nobody edits a page that was retired for a reason
Comments stay open: you can still record why a page was retired, right on the page
Point at the replacement: name a successor and the archived page links forward to it, while the live page shows what it supersedes
Whole branches at once: archiving takes the nested documents with it, and the dialog tells you how many will move before you confirm
Restore deliberately: bringing back a parent leaves its children in the archive — each was retired for its own reason
Never expires: the archive is not the trash and nothing is swept out of it
From the CLI too: meldoc archive and meldoc unarchive, plus a project setting that makes a file removed from your repo archive the page instead of deleting it
Renames That Don't Break Links
Renaming a document used to be quiet damage: links kept working for a while, then broke one at a time, months later.
Old names keep resolving: the previous alias becomes a redirect, so links, bookmarks, and synced files survive the rename — everywhere, including public pages and the CLI
Two badges, one decision: the connections panel marks a link as old name when a redirect still covers it, or will break when nothing resolves that name any more
A cleanup queue: the new Stale aliases tab collects every retired name links still use, ordered by how recently each was used — so the redirects people actually depend on rise to the top
Project renames too: renaming a project retires its old alias the same way, and the workspace-scoped view catches the links it reaches in other projects
Rewrite links in one pass: meldoc migrate moves repository files off a retired name and updates every [[link]] to it at once
Safe by refusal: deleting a document by a name that now points elsewhere is refused instead of guessed
Teams
Grant a project to a group: create a team, add members, and give the team a role on a project — everyone in it gets that access
One access list: teams and individual collaborators sit together on the project, each with the same roles
Highest right wins: a person holds the best of their personal grant and every team grant that reaches them — teams only ever raise access
See where access came from: View effective access names everyone who can reach a project and whether their role came from a team, a personal grant, or their workspace role
Clear boundaries: project admins manage their own project's team access, while editing a team's roster stays with workspace admins — a roster change reaches every project that team touches
AI Assistants
Create projects and manage access: assistants can now create a project and add existing workspace members as contributors
Permissions match the app exactly: MCP and CLI resolve project permissions through the same rules as the web app, so an assistant can never do more than you can
Filtering folded into listing: document filters now live on docs_list instead of a separate tool, with per-project permissions reported alongside each project
Archive-aware: assistants archive and restore through docs_update, and every listing hides archived pages unless asked
Stale-alias audit: docs_audit gained a check for links still using retired names — the only one that can run across a whole workspace
Search
Exact names find their document: typing a document's alias now goes straight to it, even when those words never appear in the text — and former aliases work too
Include archived: archived pages stay out of results until you ask for them
Public search matches the public tree: a hidden page no longer surfaced in search while being absent from navigation
Plans & Limits
Duplicate Radar on Pro: duplicate detection is now available from the Pro plan up
Teams on paid plans: teams are available from the Team plan up; the tab shows what you'd get on Free
Webhook limits follow your plan instead of one fixed number
Audit Log
New events: alias renames, archiving and restoring, and team access changes are now recorded
Longer memory on Enterprise: "unlimited" retention now has a definition — history is kept for years rather than trimmed to a plan window
Duplicate Radar
Templates stop clustering: near-empty documents are left out of duplicate detection, so a folder of scaffolds no longer reports itself as a pile of duplicates
Webhooks
Deletes report the whole branch: docs.changed now lists every descendant of a deleted document, not just the one you deleted
New events: doc.archived and doc.unarchived
Faster delivery: the dispatcher wakes the worker instead of polling
Security & Reliability
Invitations expire: an invite is valid for seven days, matching the link in the email, and each address has one live invite per workspace
Hardened sign-in: login codes and device codes are stored hashed, a replayed refresh token is told apart from a two-tab race, and an OAuth login no longer adopts an account on an unverified email
Trash is read-only: a document in the trash refuses edits instead of accepting changes nobody can see
Publishing fixed: the first public project of a workspace can now enter the public showcase — it never could before
Restored documents are searchable again: restoring from the trash re-indexes the document
Version History
Every document now keeps its own history, and you can read it without leaving the page.
Browse past versions: open the history panel to see each version with its author, when it landed, and how many saves it absorbed
Snapshot or diff: read a version as it was, or switch to a line-by-line diff against the live document
Restore what you need: bring back the content, the title, or the custom field values — separately, so recovering one field doesn't undo a week of edits
Sessions, not saves: rapid edits are grouped into one version instead of flooding the list with autosaves
Through your AI assistant: docs_history lets an assistant read versions and diffs, and report what changed in its own edits
Locked documents stay locked: restore is refused on a locked document, so automation-managed pages can't be rolled back by accident
Custom Fields
Default values: give a field a default and every document inherits it — change the default later and they all follow, without rewriting a single page
Deliberately empty: a cleared field now stays empty even when the default changes, so "this genuinely has no owner" is a state you can express
Rename options safely: renaming a select option migrates every document holding the old value instead of stranding them on a value that no longer exists
Order and layout: drag fields into the order you want, and choose whether they render as a table, an inline row, or a stacked list per project
Live updates: field values and definitions broadcast over websockets, so a teammate's change appears in your open document
Audited: field values and schema changes now show up in the audit log
Document Tree Filters & Sorting
Filter the sidebar: narrow the tree to the documents you're working on — by name, workflow, dates, author, or any custom field
Sort any way you like: keep your manual hierarchy, or sort by name, date, or a field value
Ask the awkward questions: "is empty" and "is not empty" cover the cases plain conditions can't — an empty parent means the document sits at the root
Inherited values count: filtering by a field also matches documents that take that value from the project default
Same filters over MCP: assistants can ask for documents by exact criteria instead of searching prose
Connect an AI Assistant
Setup wizard: pick your client — Claude, Claude Code, Cursor, VS Code, Windsurf, Codex — and get a one-click install button or a copy-ready command, with the workspace and a fresh token already filled in
Pin a workspace: a connector URL can name its workspace, so accounts with several workspaces stop guessing which one a request means
See what's connected: the workspace Overview now tracks which assistants have actually connected, and flips to "Connected" on the first request
Search & Discovery
Gap detection: repeated searches that came back empty are collected and ranked, so the holes in your documentation become a list instead of a feeling
Semantic neighbours: the connections panel now shows documents that are related in meaning, not just the ones you linked by hand
Steadier duplicate scores: a duplicate pair now reports the same similarity from either side
Team
See who's around: the workspace Overview carries a strip of your team, with a green dot on whoever is currently in the workspace
Names, not just emails: members and audit entries show names with colored initial avatars, sorted by role
Comment faster: a keyboard shortcut starts a comment on the current selection
Fixes
Websocket subscriptions are authorized: a client can no longer subscribe to channels it has no right to read
Login returns you where you were: signing in resumes the page you came from instead of dropping you on the dashboard
Stale tabs reload themselves: a tab left open across a release refreshes instead of running old code
Mobile tabs scroll: document and project overview tabs no longer overflow on narrow screens
Custom Fields
Structured metadata on documents: define custom fields per project — text, number, boolean, list, or select (single choice from options) — and fill them in from the document properties panel
Fields via MCP: AI assistants can define fields, list them, and set values on documents through dedicated MCP tools
Fields via CLI: field values sync through frontmatter, so meldoc push and meldoc pull keep local files and the workspace in step
Permissions: collaborators get read-only access to field definitions; managing fields requires project write access
Inline Comments
Comment on any selection: select text in a document and start a thread — with a keyboard shortcut for quick access
Threads: reply, edit, resolve, and reopen; a document comments panel collects all threads in one place
Comments via MCP: AI assistants can read, create, reply to, and resolve comment threads through dedicated MCP tools
Audited: comment actions appear in the audit log, including those made through MCP
Document Sharing & Short Links
Stable short links: every document gets an immutable short ID, so internal links survive renames and alias changes
Share page: share a document with your team or publish a public link from a single dialog
Clearer audience labels: exposure options now describe who can actually see the document instead of abstract level names
Graceful misses: unresolved short links land on a dedicated not-found page, and resolves are rate-limited per user
Editor
WYSIWYG tables: tables are now a visual widget — add and remove rows and columns from a table toolbar, undo/redo per cell, and scroll wide tables horizontally without breaking the layout
Unified selection toolbar: one formatting toolbar for text selections across the document, including inside tables
Figma embeds: paste a Figma link to embed the file or frame inline, with a controls overlay; plain Figma links get inline previews too
Smoother long documents: editor widgets (code blocks, images, tables) now stay alive while scrolling instead of re-rendering
Attachment previews: images in the document files panel open in a zoomable preview with a download button
MCP & Connectors
Claude.ai connector setup: a guided dialog for connecting your workspace to Claude.ai as a custom connector, integrated with token management
Workspace-scoped MCP endpoints: each workspace connector gets its own MCP mount point, with path-aware resource metadata and OAuth issuers scoped per path
OAuth Client ID support: MCP tokens can carry a Client ID for OAuth integrations, and rejected tokens are logged with detailed reasons for easier troubleshooting
Documents & Navigation
Draft toggle: choose draft or published right in the document creation dialog — new documents now default to published
Tree follows you: the document tree automatically reveals and scrolls to the active document, and the navigation menu gets a sticky search input
Status at a glance: info tooltips explain draft, locked, and hidden-from-navigation states on the document header
Search & Duplicate Radar
Project-scoped search: limit search to a single project instead of the whole workspace
Shared "not a duplicate": dismissing a duplicate pair now applies workspace-wide, so the same pair doesn't resurface for your teammates
Knowledge Graph
Three views, one graph: explore your knowledge as a workspace graph, a project graph, or a document-centred graph — switch zoom levels to see the big picture or drill into a single document's neighbourhood
Semantic edges: the graph now surfaces connections beyond explicit links — documents related by meaning appear as semantic edges, with a relevance score, even if no one linked them manually
Click-through navigation: click any node to jump to the document, including across projects
Connections panel: the document sidebar now consolidates outgoing, incoming, and external links under a single Links header with a total count, replacing the older split panels
Duplicate Radar
On-document duplicates badge: a badge on the document page surfaces near-duplicate neighbours so you can merge or differentiate before they drift further apart
Workspace-wide duplicates page: see clusters of similar documents across the whole workspace, ranked by similarity
Resilient: lazy reindex when a document hasn't been embedded yet, and graceful degradation on transient embedding errors
Broken Links Audit
Per-project broken-links widget: every project gets a widget that lists documents with broken outgoing links — fix them in place from the widget
Smart link parser: links inside fenced or inline code blocks, and escaped brackets, are no longer reported as broken
Move and delete cascade: deleting or moving a document now updates the link index automatically, so the broken-links list stays accurate
Cross-Project Document Move
Move documents between projects: from the document menu, move a document — and its entire subtree — to another project in the same workspace
Permissions: write+ access on both source and target projects is required
Editor
Fullscreen image preview: click an image to open a fullscreen dialog with zoom and pan
Fullscreen Mermaid preview: same fullscreen viewer for Mermaid diagrams — useful for large flowcharts
Heading anchors: hover any heading to copy a direct link; opening a link to a heading soft-highlights it on arrival
Glossary in tables: glossary terms are now highlighted with tooltips inside table preview cells too
Code-block polish: textarea padding aligned with the highlight layer, fixes around inline code and markdown-language blocks
MCP & AI Assistants
Fast, incremental MCP tools: the MCP server now serves documents through fast, incremental tools — partial document reads, block-level edits, and targeted audits, so AI assistants don't have to re-read or rewrite full documents
docs_audit tool: audit your knowledge base from an AI assistant — kind=broken_links for broken-link reports, plus duplicate-radar enrichment that flags near-duplicates inline in search and audit results
docs_update block-level edits: edit documents by heading path or block range with insert modes; dryRun previews the patch, and conflicts include currentUpdatedAt so assistants can resolve without re-fetching
docs_get partial reads: fetch a single section or heading range instead of the whole document
docs_search improvements: results include node degree and hub status; group results by project; alias matcher now catches paraphrased and quoted queries
docs_related views: outgoing, incoming, and suggest-links views replace the older docs_links/docs_backlinks tools
App deep links: server_info now returns the app base URL and link templates, and documents come back with ready-to-use app URLs — assistants can always hand you a clickable link to a doc, project, or the workspace
kind=context: a workspace AI playbook returned on demand so assistants always have current workspace context
MELDOC_TOKEN: the integration token env var is now MELDOC_TOKEN (replaces MELDOC_MCP_TOKEN)
Search
Heading-aware chunking: retrieval now chunks documents along heading boundaries — snippets stay coherent and on-topic
Multi-source snippets: a single search result can stitch together evidence from multiple parts of a document
Graceful degradation: search falls back to FTS-only when the workspace has no embeddings yet, instead of returning empty results
Workspace & Admin
Consolidated workspace overview: projects list and workspace navigation are unified into a single Overview page with tabs
Workspace markdown export: admins can export the entire workspace as a Markdown bundle from general settings
Embeddings reindex UI: super-admins can trigger workspace- or project-wide embeddings reindex from the admin panel, with a per-workspace embeddings health card to spot indexing problems
Audit detail: document-update entries in the audit panel now show a breakdown by patch type, so you can tell a full rewrite from a block-level edit
Sign-in & Reliability
Flexible OAuth linking: link an OAuth provider to your account even when the provider email differs from your account email
Sessions that stay put: sign-in sessions now survive tab wake-ups, concurrent token refreshes, and signup races — no more surprise logouts
Realtime resilience: online presence handles multiple tabs correctly, and workspace member lists resync automatically after a reconnect
Terminology & Glossary
Workspace Glossary: define terms once and see them highlighted with a hover tooltip across all documents in the project — add markdown descriptions, examples, and links to each term
Real-time sync: glossary changes (create, update, delete) propagate instantly to all open sessions via WebSocket — no reload needed
Import / Export: export the full glossary as a file and import it back, useful for seeding a new project or sharing terminology across workspaces
Glossary via MCP: AI assistants can read and manage glossary terms through dedicated MCP tools, including batch get, add, update, and delete operations; integration tokens get a new canGlossaryWrite permission to control write access
Plan limits: glossary term count is tracked in entitlements and visible in the plan limits page and audit log
Integrations & AI Assistants
Outbound Webhooks: subscribe to document events and receive HTTP notifications — connect Meldoc to CI pipelines, Slack, or any custom automation
MCP streaming notifications: AI assistants now receive progress and logging notifications during long operations, and a tools/list_changed notification when the server's tool set updates
Tool annotations & server instructions: MCP tools now include annotations and a server-level instruction block so AI assistants have better context on how to use them
OAuth discovery: /mcp now returns 401 + WWW-Authenticate for unauthenticated requests, making it easier for OAuth-aware clients to discover the authorization endpoint automatically
Onboarding
Demo project: new workspaces automatically receive a ready-made demo project — a quick way to explore the editor, document tree, and glossary without starting from scratch
Search
Hybrid search: queries now combine full-text search and embeddings for more relevant results; if embeddings are unavailable, the engine falls back to FTS-only automatically
Editor
Code-block improvements: better syntax rendering for multi-language blocks; various edge-case fixes
List and Enter-key fixes: list marker alignment and Enter-key continuation behavior corrected for nested lists
Double-click word selection: double-clicking selects the word under the cursor, matching standard editor behavior
Magic links in table cells: magic link widgets now work inside table cells
Admin & Reliability
Super-admin tools: new activity stats dashboard and error-log management interface; a button to broadcast MCP tools/list_changed to all connected clients
Seamless deploys: new chunks from a deploy are retained alongside the previous version — users see a soft "update available" banner instead of a broken page after a release
Auth resilience: refresh cookies are sent on all API calls; Bearer tokens sync from X-New-Access-Token; silent-refresh errors no longer clear the refresh cookie; failed requests are automatically retried with the new token after a 401
CLI
meldoc ci command: a dedicated command for CI/CD pipelines — validates and publishes docs in a single step with machine-readable output
Pre-publish validation: meldoc push now runs validation before publishing; checks include content links, cross-project references, alias cycles, and field completeness; use --fix to auto-resolve simple issues or --skip-validate to bypass in trusted pipelines
Hybrid proxy MCP server: the local MCP server now proxies remote tools dynamically — tool definitions are fetched at runtime rather than hardcoded
Skills management: meldoc skills commands for listing, adding, and removing workspace skills
Integrations & AI Assistants
Integration tokens: a single access token now works across the CLI, AI assistants, and all integrations — no more creating separate tokens per tool. Manage tokens directly from project settings
New MCP transport (Streamable HTTP): AI assistant connections (Claude, Cursor, etc.) now use Streamable HTTP with OAuth 2.1 — more reliable than the previous SSE transport, with proper session handling
Device flow shows your account: when authorizing an AI assistant via the device flow, you now see the email of the authenticated account before confirming — so you always know which workspace you're connecting to
Editor
Document duplication: duplicate any document directly from the document tree or documentation page — opens a pre-filled creation dialog with the content ready to copy
Image widget improvements: delete button for removing images, hover effects for empty state, support for external URL images, CDN-backed asset loading
Access Control
Permission-based document deletion: the delete action in the document tree and documentation page is now only available to users with delete permissions — reflects project-level collaborator roles
Collaborator management: users and integration tokens are now listed separately in project settings — easier to see who and what has access to your project
Assets
CDN asset delivery: assets are now served via CDN using SHA256-based URLs for better performance and cache reliability
Onboarding & Billing
Plan selection on signup: new workspaces choose a plan during creation and go through Stripe Checkout directly — no extra step after signup
Subscription self-service: cancel or reactivate a subscription from workspace settings; access the Stripe billing portal for payment details and invoices
Seat limits on invites: the invite flow enforces seat limits by plan — admins see clear errors when their plan is full
Workspace blocked banner: when a workspace is in a payment-required state, the banner now includes a direct action to resolve it and updates in real time via WebSocket
Auth
Long-lived refresh tokens: OAuth refresh tokens now have a longer lifetime with proper rotation — fewer unexpected logouts for long-running sessions
MCP Package
Lightweight MCP package: introduced a standalone npm package @meldocio/mcp-stdio-proxy for MCP integration
Easy installation via npx: npx -y @meldocio/mcp-stdio-proxy@latest
No need to install the full CLI — use MCP functionality independently
Automatically configured when running meldoc mcp init for IDE setups (Cursor, VS Code, Windsurf, Claude Code)
Works seamlessly with all MCP-compatible tools and IDEs
Authentication & Authorization
Device-code authorization: device-based authentication through MCP proxy — no separate CLI or MCP tokens needed
Login through MCP proxy opens browser for authentication
Tokens stored in ~/.meldoc/credentials.json and automatically refreshed
Authenticate once, use across all contexts on the device
Still supports MELDOC_ACCESS_TOKEN for CI/CD and integrations
CLI
push command: meldoc publish is now meldoc push — shorter and more consistent. The old publish still works but shows a deprecation warning
meldoc new: create a new local .meldoc.md file with frontmatter — interactively or via flags (--alias, --title, --parent). Also available as MCP tool cli_new
meldoc docs: list all local documentation files with hierarchy and frontmatter. Detects duplicate aliases. Also available as MCP tool cli_docs_list
meldoc organize: reorganize local documentation files into a folder structure that matches their hierarchy. Supports --dry-run for safe previewing. Also available as MCP tool cli_organize
Document ordering: add an order field to frontmatter to control the position of a document among its siblings. Duplicate values under the same parent are flagged as warnings
Scope filtering: scope and scopeChildren options in .meldoc.yaml to limit which documents are included in push/pull operations — useful for partial documentation setups
meldoc updateVersion: update the CLI binary to the latest release without reinstalling
Loading indicators: CLI commands now show spinners during long-running operations
App
Project collaborators: invite team members with specific roles at the project level, independent of their workspace-wide role. Manage collaborators in project settings
Project overview: workspace slug and project alias now visible in the project card, with one-click copy
Search
Improved search relevance: prefix search strategy for better partial-word matching; results now boost documents whose title or alias matches the query
AI Writer
Quick AI commands: rewrite, improve, fix grammar, simplify, summarize, explain, translate, change tone, make shorter/longer, continue writing — all available directly in the editor
Automatic document linking: AI Writer suggests and inserts MagicLinks based on aliases and documentation structure
Editor & Authoring Experience
Syntax highlighting for code blocks: code snippets in the editor now display with proper language-specific highlighting
Mermaid diagrams support: create flowcharts, sequence diagrams, and more using Mermaid syntax
Video assets support: upload and embed MP4 and WebM videos directly in your documentation
External video embedding: embed YouTube and Vimeo videos using simple markdown image syntax
AI-Friendly Documentation
MCP server support: connect AI tools (Cursor, Claude, etc.) directly to your documentation
Works with both public and private documentation — all access is secured via client tokens
AI assistants can now read your docs without exposing source code or repository access
Great for product teams, support, integrators, and anyone who needs reliable context without digging through code
MCP HTTP transport (SSE): streamable HTTP transport support for MCP connections
MCP tools for document management: create and delete documents directly via MCP
MCP in meldoc CLI: added MCP server for IDE agents and Claude Code via the CLI
CLI tools for MCP: create and delete documents via MCP server in CLI
CLI document deletion: delete documents directly via CLI commands
.mcp.json support: configure MCP servers using standard .mcp.json configuration files
llms.txt support: public documentation now includes llms.txt at workspace, project, and document levels
Improves discoverability and context quality for AI agents and LLM-based tools
Activity & Transparency
Document-level activity views: see the full history of changes for any document
Project-level activity views: track activity across all documents in a project
Token attribution in activity logs: see exactly which CLI or MCP token made each change
Expanded visibility options: admins can now monitor changes at workspace, project, and document levels
Public App
Syntax highlighting: code blocks in public documentation now display with proper syntax highlighting
Theme switching: readers can toggle between light and dark themes in public documentation
Real-time Presence
Show who is currently viewing: see which team members are viewing the same document in real time
WebSocket-based presence: instant status updates for collaborative awareness
Collaboration & Availability
Smart document synchronization: documents automatically sync when the editor loses focus or after 30 seconds of inactivity, ensuring changes are saved without interrupting your workflow
Access Control
Password-protected documents: secure public documents with password access
MCP token permissions: granular permission control for MCP tokens (read-only, create, update, delete)
CLI token permissions: granular permission control for CLI tokens (read-only, create, update, delete)
Integrations
CI/CD integrations: automate documentation workflows with GitHub Actions and other CI/CD pipelines
Assets & Media
CLI ↔ UI asset workflow: Work with images (extensible for other file types)
Asset binding: Assets as first-class citizens of documentation
Draft Documents
Explicit draft status for documents
Block publishing drafts via CLI
Exclude drafts from public viewer
Optional exclusion from AI / semantic search
Activity & Transparency
Added workspace activity log: see who changed what and when
Activity visible at workspace level (currently for admins)
Laid foundation for document- and project-level activity views
Improved trust, collaboration, and accountability for team documentation
Editor & Reading Experience
Inline document linking widget: Search and insert document links directly from the editor
Improved Markdown editor built on custom CodeMirror setup
Clear separation between editing and reading modes
Added navigation by headings (TOC) for long documents
Added recent documents view in projects
Added command palette for keyboard-driven editor actions
Public Documentation Sharing
Navigation by headings (TOC) in public documents
Manage public project order: configure your public workspace storefront
Publish entire projects or individual documents
Full control over what is visible outside your workspace
Read-only public access without login required
Optimized for fast, stateless delivery
Ideal for sharing specs, docs, or architecture notes
Works well for training and grounding third-party AI agents
AI Assistance
Context-aware writing and editing: improve selected text or generate new content
In-editor knowledge lookup: search across documentation while writing
Summarization and translation of documents or sections
AI adapts to prompt and chooses appropriate action based on documented knowledge
Ask AI: query-aware semantic search with natural language questions
Multi-step document retrieval with automatic re-querying
Answer-oriented results with summaries and source document links
Knowledge Connections
Added instant cross-links between documents using simple references
Improved navigation in complex knowledge bases
Product & Monetization
Product plans: Free / Pro / Enterprise
Organization-level feature flags
Clear feedback and upgrade flow for paywalls
CLI Improvements
More reliable context handling when switching between projects and workspaces
Added migration commands to migrate documentation between CLI versions
Added validation tools to detect issues before publishing
Agent instructions support: automatic creation and updates of instructions file for AI agents
Magic links processing: CLI detects and processes links between documents
Improved stability for CI and team workflows
CLI
Enhanced frontmatter parsing and metadata handling
More accurate SHA256 hashing
Improved publish and sync logic (publish / pull)
More reliable conflict resolution with --force support
Improved error handling in pull operations
Added debug mode
Added MELDOC_TOKEN to init command
Added help command
Improved CLI installation process
Search & Navigation
Improved semantic search: more relevant results
Added text previews in search results
Improved navigation between documents
Document Management
Added automatic document locking when editing
Improved project and document publishing management
Removed legacy public routes, fully replaced with secure publishing logic
Cloud Workspace & Teams
Full invitation lifecycle: send, update roles, decline, resend
Improved role and permission UI components
Updated welcome and notification email templates
AI
AI text improvement integrated into editor
Improved loading state display and event handling
UI & UX
Improved typography and theme
Improved mobile navigation
Updated branding elements
CLI
Improved manifest handling
Improved cursor handling for pull
Simplified API fields in CLI
Improved SHA256 hashing
Improved installation scripts
Improved installation guide (TOC, structure)
Documentation UI
Added public documentation mode
Added lock/unlock component for documents
Improved mobile documentation version and navigation tree
Clients (CLI Tokens)
Full UI for managing CLI clients (API tokens)
Branding
Updated OAuth messages
Updated favicon
Search
Improved connection between search results and projects
Improved error handling
CLI
Improved state management and indexes
Improved doc_id handling
Improved file scanning logic
Updated CLI documentation and .meldoc directory
Cloud
Added public documentation routing
Added anonymous read-only mode
Workspace & Navigation
Improved nested navigation
Improved module routing
Auth
Improved social login
Improved token-login
Editor & Projects
Added project description
Improved document metadata handling
CLI — Initial Release Foundations
meldoc init — initialize project
meldoc scan — find .meldoc.md files
meldoc publish — publish documents to cloud
meldoc pull — sync documentation
Full YAML frontmatter support
SHA256 file hashing
Conflict resolution (--resolve)
Installers for Linux, macOS, Windows
Token-based authentication
Multi-platform builds and releases
Cloud
Created workspace / project / document entities
Added core APIs
Added role model (owner/editor/viewer)
Basic document storage
Editor & Viewer (v0)
First Markdown renderer
First editor prototype
Routing & Navigation
Basic route: workspace → project → document
Brand
Initial version of Meldoc branding
Added base color tokens and typography
Cookie preferences
We use Google Analytics to see which pages are read and which buttons are used. Nothing is sent to Google
and no cookies are set unless you accept. You can change your choice any time under Cookie settings at the bottom of the page.
Cookie policy
Cookie settings
Essential cookies
Always active
Needed for the site to work, so always on. On meldoc.io this is only the cookie that remembers your choice.
Analytics cookies
Google Analytics shows us which pages are read and which buttons are used. It keeps a random identifier in
a cookie to tell visits apart; we send it no names, email addresses or document contents. Meldoc uses no
advertising cookies.