Is there an MCP server that serves a repo's docs to coding agents?
Yes — trovex is an MCP server built for this. It indexes a repo's markdown and returns the single current doc that answers a query — a path:line pointer with a freshness marker — instead of letting the agent reread the repo or sift a pile of candidate chunks. It serves the section that answers for about 60% fewer tokens per lookup, runs locally, and works with any MCP client.
What makes it different from a generic context server
Most retrieval servers hand back the top-k similar chunks and leave the agent to rank them, with no freshness signal. trovex is opinionated for repo docs: it returns one canonical doc, marks stale and duplicate copies so they're skipped, reads at the section level, and closes the write loop — an agent records what it learns once and the rest read it back. See trovex vs context-hub and vs a vector-DB RAG setup for the contrast.
How it runs
trovex serves MCP over streamable HTTP at http://localhost:8765/mcp. Indexing and embeddings run on your machine — vectors in SQLite, ONNX embeddings — so there's no cloud and no API keys, and your code never leaves your machine. Connect it from Claude Code, Cursor, Windsurf, Cline, or any MCP client.
An MCP server for your repo's docs.
trovex is in private beta — request access and connect it to your agent.
Open source (AGPL-3.0 core, MIT CLIs). Local-first — your docs never leave your machine.