rock-mcp

rock-mcp

Documentation crawler MCP server. Share a link → site gets crawled and indexed → any MCP-compatible AI can search, read, and expand on it.

No API keys. No cloud. Runs locally, stores everything in SQLite.

How it works

1. Share a docs URL (add_docs_url)
2. rock-mcp crawls and indexes it in the background
3. Ask your AI about it — it searches, reads pages, and expands topics on demand

Tools

| Tool | Description | |---|---| | search_web | Search the web via DuckDuckGo — find URLs when you don't have one yet | | add_docs_url | Crawl and index a URL (async, call immediately then search) | | search_docs | Full-text search across all indexed docs | | get_page | Get a specific page, chunked. Jump to a section by name. | | expand_topic | Find all pages covering a topic | | list_sources | List indexed sources with crawl status and staleness | | recrawl_source | Re-crawl a source to pick up updates | | get_crawl_status | Check crawl progress for a URL | | delete_source | Remove a source and all its pages |

Installation

Prerequisites

• Node.js 18+
• Git

git clone https://github.com/sethwebster/rock-mcp
cd rock-mcp
npm install && npm run build

Note the full path to dist/server.js — you'll need it below.

---

Claude Code (CLI)

claude mcp add --scope user rock-mcp node /path/to/rock-mcp/dist/server.js

Restart Claude Code after running this. Verify with:

claude mcp list

---

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "rock-mcp": {
      "command": "node",
      "args": ["/path/to/rock-mcp/dist/server.js"]
    }
  }
}

Restart Claude Desktop.

---

Cursor

Add to .cursor/mcp.json in your project root, or ~/.cursor/mcp.json globally:

{
  "mcpServers": {
    "rock-mcp": {
      "command": "node",
      "args": ["/path/to/rock-mcp/dist/server.js"]
    }
  }
}

Restart Cursor.

---

Windsurf

Add to ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "rock-mcp": {
      "command": "node",
      "args": ["/path/to/rock-mcp/dist/server.js"]
    }
  }
}

Restart Windsurf.

---

Codex (OpenAI CLI)

Add to ~/.codex/config.yaml (or your project's codex.yaml):

mcp_servers:
  - name: rock-mcp
    command: node
    args:
      - /path/to/rock-mcp/dist/server.js

---

Cline (VS Code extension)

Open VS Code settings, search for Cline MCP, and add:

{
  "rock-mcp": {
    "command": "node",
    "args": ["/path/to/rock-mcp/dist/server.js"]
  }
}

Or edit settings.json directly under the cline.mcpServers key.

---

Continue.dev

Add to ~/.continue/config.json under mcpServers:

{
  "mcpServers": [
    {
      "name": "rock-mcp",
      "command": "node",
      "args": ["/path/to/rock-mcp/dist/server.js"]
    }
  ]
}

---

Antigravity

In the agent panel, click ... → Manage MCP Servers → View raw config to open mcp_config.json. Add your server:

{
  "mcpServers": {
    "rock-mcp": {
      "command": "node",
      "args": ["/path/to/rock-mcp/dist/server.js"]
    }
  }
}

Save the file and restart Antigravity (or reload the window).

---

Firebase Studio (Project IDX)

Add to .idx/mcp.json in your workspace root (create it if it doesn't exist):

{
  "mcpServers": {
    "rock-mcp": {
      "command": "node",
      "args": ["/path/to/rock-mcp/dist/server.js"]
    }
  }
}

For Gemini CLI within Firebase Studio, add to .gemini/settings.json instead.

---

Zed

Add to ~/.config/zed/settings.json under context_servers:

{
  "context_servers": {
    "rock-mcp": {
      "command": {
        "path": "node",
        "args": ["/path/to/rock-mcp/dist/server.js"]
      }
    }
  }
}

---

Any MCP-compatible client

rock-mcp is a standard MCP stdio server. The binary is node /path/to/rock-mcp/dist/server.js. If your client supports MCP, point it there.

---

Storage

All data lives in ~/.rock-mcp/docs.db (SQLite, WAL mode). Delete it to reset everything.

Crawl defaults

• Depth: 2 hops from the root URL
• Max pages: 50 per source
• Concurrency: 5 pages at a time, 200ms between batches
• Only indexes text/html — skips PDFs, plain text, JSON, etc.
• Redirect deduplication: follows redirects and stores under the canonical URL

License

MIT