Two people. Two projects. Two pages about the same thing — with different details and different degrees of staleness. Nobody is at fault: that’s how any knowledge base grows once more than one person writes in it.

Why duplicates are dangerous

A duplicate isn’t just clutter. It’s two versions of the truth that drift apart with every update: one got fixed, the other got forgotten.

A reader finds the stale copy — and is sure they’ve read the documentation. An AI agent finds both — and assembles an answer from contradicting sources.

The classic cure is the “deep clean”: once a quarter someone heroically re-reads the base. By the next quarter the duplicates have grown back.

How it works in Meldoc

Meldoc compares documents at the level of meaning, not text. Every document has a semantic fingerprint — a mechanism similar to the third layer of the graph (Graphs in Meldoc). The system finds close pairs ahead of time: the signal lives where agents work with documents — in search through MCP. To humans, Meldoc shows duplicates differently: as a marker on the document itself and in the workspace-wide duplicates overview. An agent working through MCP (MCP in Meldoc) sees the signal “these two results are probably duplicates” right in the data it works with — no separate audit needed.

False positives are manageable: you can mark a pair as “not a duplicate,” and the system will stop suggesting it.

What to do with what you find

There are three options. Merge the two documents into one. Keep one canonical and link to it from the other with a magic link. Or hand the task to an agent: through MCP it will read both, merge them and place the links itself.

The takeaway

You don’t have to scrub the knowledge base on Saturdays. Duplicates are an inevitable byproduct of living documentation: you can catch them one at a time, the moment you’re looking at them, or you can sit down and sweep the whole project in one call (“We should clean up the docs”). Either way, Meldoc simply makes the duplicates visible.