Open any MCP directory and count the connectors. Slack. GitHub. Google Drive. Notion. Linear. Postgres. Every single one does the same thing under a different name: it gives your AI a way to fetch something.
That's useful. It's also the ceiling. Fetching a file is not the same as understanding why the file exists, what it replaced, or what broke when someone ignored it.
Here's a concrete example. Say you connect a Drive MCP and ask Claude "why do we rate-limit this endpoint at 50 req/s instead of 200?" The Drive MCP can search for documents containing "rate limit." If you're lucky, it finds a doc. That doc says "50 req/s, see incident review." Now what? The MCP has no way to follow that thread. It found a file. It did not find the reason.
The reason usually lives in three places at once: an incident writeup, a Slack thread nobody archived, and the head of whoever was on call that week. Search retrieves objects. It doesn't retrieve relationships. And "why" questions are, almost by definition, questions about relationships between things, not about any single thing.
This is the gap docs-MCP servers are supposed to close, except almost nobody has actually built one. What exists today is "expose a file store over MCP." What doesn't exist is "expose the graph of how your team's documents relate to each other, and let the model walk that graph to answer why, not just what."
We built the second one. Rootr keeps your team's docs, decisions, and postmortems as a live workspace, and it already builds a knowledge graph out of the connections between them: this doc supersedes that one, this decision references that incident, this spec was revised because of that customer conversation. When you connect it over MCP, the model isn't fetching a document. It's asking the graph a root-cause question and getting an answer assembled from however many documents that answer actually requires.
Practically, the difference shows up in three question types that a plain file-fetch MCP just can't handle well:
- "Why did we choose X over Y": needs the decision doc plus whatever came before it
- "What changed since the last time someone asked this": needs a timeline across documents, not one document
- "Is this still the current approach, or did it get superseded": needs to know which doc wins when two disagree
None of those are search problems. They're graph problems, and until you have a graph, your AI answers them by guessing or by making something up that sounds plausible.
Setup is a URL and an OAuth login, no API key to copy around: rootr.io/mcp. If you're already deep in MCP and tired of every connector doing the same fetch-and-forget trick, this is the one built for the other kind of question.
More on Rootr: rootr.io
