Doctree Mcp
Give coding agents fast BM25 search and tree navigation over your markdown docs without embeddings or extra LLM calls.
Overview
io.github.joesaby/doctree-mcp is a MCP server for the Build phase that BM25-searches and tree-navigates markdown documentation for agents without embeddings.
What is this MCP server?
- BM25 full-text search over local or repo markdown—no embeddings
- Tree navigation across doc hierarchy for staged agent reads
- DOCS_ROOT required; optional DOCS_GLOB (default **/*.md)
- DOCS_ROOTS supports multiple weighted collections
- Optional glossary.json via GLOSSARY_PATH for query expansion
- Server version 1.0.1
- Default DOCS_GLOB pattern **/*.md
- No embeddings and no LLM calls in retrieval path
Community signal: 1 GitHub stars.
What problem does it solve?
Agents waste tokens guessing doc locations when your knowledge base is plain markdown and you do not want a vector database pipeline.
Who is it for?
Indie projects with markdown-first docs in git who want cheap, offline-friendly agent retrieval over stdio.
Skip if: Teams needing semantic search across PDFs, Confluence, or massive multilingual corpora where embeddings are mandatory.
What do I get? / Deliverables
After pointing DOCS_ROOT at your repo, your agent can search and traverse docs locally with deterministic BM25 results and optional glossary expansion.
- BM25-ranked search results over configured markdown globs
- Hierarchical doc tree exploration for targeted agent reads
- Optional multi-root weighted indexes via DOCS_ROOTS
Recommended MCP Servers
Journey fit
Documentation-backed builds sit in Build: agents need reliable retrieval while you implement features and maintain internal or product docs. Docs subphase fits because the server indexes markdown repositories and exposes search plus hierarchical navigation for agent grounding.
How it compares
MCP doc retrieval server, not an LLM skill or hosted documentation SaaS.
Common Questions / FAQ
Who is io.github.joesaby/doctree-mcp for?
It is for developers and agent users who store docs as markdown and want BM25 search plus tree navigation exposed as MCP tools.
When should I use io.github.joesaby/doctree-mcp?
Use it while building or maintaining docs-heavy products whenever agents must ground answers in your on-disk markdown without embedding pipelines.
How do I add io.github.joesaby/doctree-mcp to my agent?
Register the doctree-mcp npm package with stdio (runtime hint bunx), set DOCS_ROOT to your markdown root, and reload your MCP client.