anthropic-docs-mcp

MCP server that indexes Anthropic's documentation into a searchable SQLite database. Gives Claude Code instant access to platform docs, API reference, Claude Code docs, and blog posts — no web fetching needed.

Install

Prerequisites

• Node.js 18+
• Claude Code

1. Clone and build

git clone https://github.com/giovicordova/anthropic-docs.git
cd anthropic-docs
npm install
npm run build

2. Add to Claude Code

Open your Claude Code MCP config:

code ~/.claude/.mcp.json
# or: nano ~/.claude/.mcp.json

Add the server (replace /path/to/ with where you cloned the repo):

{
  "mcpServers": {
    "anthropic-docs": {
      "command": "node",
      "args": ["/path/to/anthropic-docs/dist/index.js"]
    }
  }
}

3. Restart Claude Code

The server will automatically crawl and index all docs on first run. This takes about a minute. After that, the index refreshes daily in the background.

Tools

Tool	What it does
search_anthropic_docs	Full-text search across all docs with BM25 ranking. Filter by source.
get_doc_page	Get the full markdown of a specific page by path. Supports fuzzy matching.
list_doc_sections	Browse all indexed pages grouped by source. Filter by source.
refresh_index	Trigger a manual re-crawl if you need the latest
index_status	Check index health: page counts, staleness, crawl state per source

How it works

Claude Code <-> stdio <-> MCP Server <-> SQLite FTS5 Database
                              |
                         Background crawler
                              |
               platform.claude.com/llms-full.txt
               code.claude.com/docs/llms-full.txt
               anthropic.com/sitemap.xml → blog HTML

• Platform docs and API reference are crawled from platform.claude.com/llms-full.txt
• Claude Code docs come from code.claude.com/docs/llms-full.txt
• Blog posts from /news, /research, /engineering are crawled weekly from anthropic.com/sitemap.xml (incremental — only new posts)
• Pages are split into sections at h2/h3 headings for precise search results
• BM25 ranking weights title matches 10x and headings 5x over content
• Crawls are atomic — the old index stays intact until a new crawl completes successfully
• Docs refresh daily; blog refreshes weekly. Call refresh_index anytime for an immediate update
• Database lives at ~/.claude/mcp-data/anthropic-docs/docs.db

License

ISC