arabold/docs-mcp-server

What is Docs MCP Server?

Grounded Docs MCP Server is an open‑source tool that builds a searchable, version‑specific index of documentation for any code library or framework you use. It pulls the latest docs from official sites, GitHub repos, npm / PyPI packages, local folders or zip archives, converts them to plain text/markdown, and (optionally) stores vector embeddings so a language model can retrieve the exact passage it needs. The server runs locally, exposing a simple MCP (Message‑Stream‑Endpoint) API that AI coding assistants such as Claude, Cline, Copilot, Gemini‑CLI, etc., can query.

---

Why it matters

• Reduces hallucinations – the LLM answers from the real, up‑to‑date docs instead of guessing.
• Version‑aware – you can index the exact version of a library that your project depends on.
• Private & local – all processing stays on your machine; no code or docs leave your network.
• Broad format support – PDFs, Office files, markdown, Jupyter notebooks, and >90 source‑code languages are parsed automatically.
• Plug‑and‑play for existing AI tools – works with any MCP‑compatible client (Claude, Cline, VS Code extensions, etc.).

---

Core features (as described in the README)

Feature	Description
CLI‑first workflow	scrape, search, fetch-url commands let scripts or agents index docs, run semantic search, and retrieve raw pages.
Web UI	Accessible at http://localhost:6280; lets you add docs, view the index, and manage settings through a browser.
MCP server	Long‑running SSE endpoint (/sse) that AI clients can point to in their configuration files.
Embedding support	Optional vector embeddings (OpenAI, Ollama, Gemini, Azure, etc.) improve semantic search quality.
Docker image	Official container (ghcr.io/arabold/docs-mcp-server:latest) for isolated deployment.
Hash‑routed SPA handling	--preserve-hashes flag (or UI checkbox) tells the scraper to keep URL fragments and, when needed, fall back to Playwright for client‑side rendering.
Markdown‑first crawling	Sends Accept: text/markdown header; if the site can serve markdown directly the scraper skips HTML‑to‑markdown conversion.
Extensive format list	PDFs, Word/Excel/PowerPoint, OpenDocument, RTF, EPUB, Jupyter notebooks, archives, HTML, markdown variants, many programming languages, config files, data formats, etc.
Authentication & security	Docs mention OAuth2/OIDC support and hardening guides for production deployments.

---

Quick‑start guide (from the README)

1. Install / run – No permanent install needed; use npx (Node 22+):

   npx @arabold/docs-mcp-server@latest          # starts the server & UI
   

   Or run the Docker container:

   docker run --rm -v docs-mcp-data:/data -v docs-mcp-config:/config \
     -p 6280:6280 ghcr.io/arabold/docs-mcp-server:latest \
     --protocol http --host 0.0.0.0 --port 6280
   

2. Index a doc source (CLI example):

   npx @arabold/docs-mcp-server@latest scrape react https://react.dev/reference/react
   

   Add --preserve-hashes for hash‑based SPA docs.
3. Search the index:

   npx @arabold/docs-mcp-server@latest search react "useEffect cleanup" --output yaml
   

4. Connect an AI client – add the server to the client’s MCP config, e.g. for Claude Desktop:

   {
     "mcpServers": {
       "docs-mcp-server": { "type": "sse", "url": "http://localhost:6280/sse" }
     }
   }
   

5. (Optional) Enable embeddings – set OPENAI_API_KEY (or other provider env vars) before starting the server to get semantic search.

---

Who might use it?

• Developers building AI‑assisted coding tools that need reliable reference material.
• Teams that want a private, version‑locked knowledge base for internal libraries.
• Researchers experimenting with retrieval‑augmented generation (RAG) pipelines.
• Tool makers who need a drop‑in MCP endpoint for Claude, Cline, Copilot, Gemini‑CLI, etc.

---

Where to find more information

• Installation & setup – docs/setup/installation.md
• Connecting clients – docs/guides/mcp-clients.md
• Embedding model configuration – docs/guides/embedding-models.md
• Supported formats list – docs/concepts/supported-formats.md
• Architecture overview – ARCHITECTURE.md
• Contribution guide – CONTRIBUTING.md

---

License

MIT – free to use, modify, and redistribute.