Docs MCP Server for Claude (Grounded Docs)

Overview

The Grounded Docs MCP Server gives Claude access to an always-current, version-specific documentation index that you control. Start the server, use the web UI or CLI to scrape documentation from any source — websites, GitHub repositories, npm/PyPI packages, or local folders — then ask Claude questions grounded in the exact library versions in your project. Licensed under MIT with 1400+ stars on GitHub.

A fully self-hosted, open-source alternative to Context7, Nia, and Ref.Tools.

Key capabilities

• Multi-source indexing — scrape websites, GitHub repos, npm packages, PyPI packages, local folders, and ZIP archives into a searchable local index.
• Version-specific search — queries target a specific library and version, so Claude answers based on the exact docs you indexed rather than training-data generalizations.
• Broad format support — HTML, Markdown, MDX, reStructuredText, PDF, Word, Excel, PowerPoint, OpenDocument, EPUB, Jupyter Notebooks, 90+ source-code languages, JSON, YAML, CSV, and more.
• CLI mode — use npx @arabold/docs-mcp-server@latest scrape react https://react.dev and search commands in pipelines and agent scripts without running the server.
• Web UI — built-in management interface at http://localhost:6280 for browsing indexed sources.
• Docker — run as a container with a persistent volume for the index.

Tools

How it compares

Grounded Docs is the only option that runs fully offline, supports Office documents, PDFs, and Jupyter Notebooks, and gives you complete control over which exact docs version is indexed.

Installation

Step 1: Start the Grounded Docs server

npx @arabold/docs-mcp-server@latest

The server starts at http://localhost:6280. Open that URL in your browser to manage indexed documentation via the web UI.

Step 2: Connect Claude Code

claude mcp add --transport http docs-mcp http://localhost:6280/sse

Step 2 (alternative): Claude Desktop

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

Docker

docker run -d -p 6280:6280 \
  -v docs-mcp-data:/data \
  ghcr.io/arabold/docs-mcp-server:latest

CLI (no server required)

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

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

Requirements

• Node.js 22+ with npx.
• An MCP client (Claude Code or Claude Desktop).
• The server must be running at http://localhost:6280 before connecting.

Security

• Fully local — no network traffic to external APIs.
• All data stored in a local SQLite database; only scrape_documentation makes outbound requests to fetch the source URLs you specify.

Source Verification Notes

Verified on 2026-06-18:

• Official GitHub repository arabold/docs-mcp-server (MIT, 1400+ stars) documents the @arabold/docs-mcp-server npm package, SSE endpoint at http://localhost:6280/sse, the five tools (scrape, search, list, remove, fetch-url), CLI mode, Docker image, Node.js 22+ requirement, and all supported file formats.
• Claude Code MCP documentation at code.claude.com/docs/en/mcp describes the --transport http SSE connection pattern used above.