A self-hosted MCP (Model Context Protocol) server that indexes your documentation and makes it searchable by AI coding assistants. Built primarily as a local-first tool — runs as a single Docker / Podman container on your workstation, with a built-in web UI for managing sources.
- Hybrid search — combines BM25 full-text search and semantic vector search (via
all-MiniLM-L6-v2) merged with reciprocal rank fusion - Multiple source types — public web docs, GitHub Wikis, GitHub repositories, Azure DevOps Wikis
- Authenticated sources — user-supplied fine-grained PATs for GitHub, Microsoft device code flow for Azure DevOps (no app registration required)
- Scheduled crawling — cron-based re-indexing keeps docs fresh
- Web UI — manage sources, trigger crawls, and test searches from a browser
- MCP tools —
search_docs,list_sources,browse_source,get_page - Zero external dependencies — SQLite with FTS5 + sqlite-vec, ONNX model bundled in the container image
The fastest way to get running — single user, container on localhost, no config file or API key needed.
Bash command
docker run -d \
--name documcp \
-p 8080:8080 \
-v documcp-data:/app/data \
ghcr.io/mathwro/documcp:latestPowerShell command
docker run -d `
--name documcp `
-p 8080:8080 `
-v documcp-data:/app/data `
ghcr.io/mathwro/documcp:latestOpen http://localhost:8080. Add a public web source from the UI and trigger a crawl. The documcp-data named volume preserves your indexed content and any tokens you store across restarts.
If you previously started this container with DOCUMCP_API_KEY set and want to return to local-only unauthenticated access, remove that environment variable and recreate the container. docker stop / docker start keeps the old container environment:
docker rm -f documcp
docker run -d \
--name documcp \
-p 8080:8080 \
-v documcp-data:/app/data \
ghcr.io/mathwro/documcp:latestFor private sources, exposing the port to your LAN, declarative source seeding via config.yaml, or building from source, see docs/install.md.
Give this to your AI coding assistant:
Please install and configure DocuMcp for my current AI tool according to this guide:
https://github.com/mathwro/DocuMcp/blob/main/docs/for-agents.md
| Topic | Page |
|---|---|
| Installation, updating, when to enable the API key | docs/install.md |
config.yaml fields and environment variables |
docs/configuration.md |
| Source types (web, GitHub Wiki, GitHub Repo, Azure DevOps) | docs/sources.md |
| Connecting Claude Desktop / Claude Code, MCP tools reference | docs/mcp-clients.md |
| AI assistant setup instructions | docs/for-agents.md |
| Common problems and how to diagnose them | docs/troubleshooting.md |
| Build, test, project layout | docs/development.md |
See CONTRIBUTING.md for the development loop, code style, and what I'm likely to accept or push back on.
To report a vulnerability, see SECURITY.md.
GPL-3.0 — see LICENSE.