MCP Server¶

RavenRustRAG includes an MCP (Model Context Protocol) server for integration with AI assistants like Claude, Copilot, Cursor, and Aider.

Starting¶

ravenrag mcp --db ./raven.db

The MCP server communicates over stdio using JSON-RPC 2.0.

Filtering Tools¶

Restrict which tools are exposed using --filter:

# Only expose search and prompt tools (read-only access)
ravenrag mcp --filter search,get_prompt

Integrations¶

1. Claude Desktop¶

Add to your Claude Desktop config file:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json
• Linux: ~/.config/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "ravenrag": {
      "command": "ravenrag",
      "args": ["mcp", "--db", "/path/to/raven.db"]
    }
  }
}

To restrict tools (read-only):

{
  "mcpServers": {
    "ravenrag": {
      "command": "ravenrag",
      "args": ["mcp", "--db", "/path/to/raven.db", "--filter", "search,get_prompt"]
    }
  }
}

2. Claude Code¶

Claude Code discovers MCP servers from the same config as Claude Desktop, or you can add one directly via CLI:

claude mcp add ravenrag -- ravenrag mcp --db /path/to/raven.db

To verify it's registered:

claude mcp list

3. Cursor¶

Add to your Cursor MCP settings:

• macOS: ~/.cursor/mcp.json
• Windows: %USERPROFILE%\.cursor\mcp.json
• Linux: ~/.cursor/mcp.json

{
  "mcpServers": {
    "ravenrag": {
      "command": "ravenrag",
      "args": ["mcp", "--db", "/path/to/raven.db"]
    }
  }
}

Or per-project in .cursor/mcp.json at the workspace root:

{
  "mcpServers": {
    "ravenrag": {
      "command": "ravenrag",
      "args": ["mcp", "--db", "./raven.db"]
    }
  }
}

4. Aider¶

Aider supports MCP servers via its --mcp flag:

aider --mcp "ravenrag mcp --db ./raven.db"

Or add to your .aider.conf.yml:

mcp-servers:
  - command: ravenrag
    args: ["mcp", "--db", "./raven.db"]

5. VS Code (GitHub Copilot)¶

Add to .vscode/mcp.json in your workspace root:

{
  "servers": {
    "ravenrag": {
      "command": "ravenrag",
      "args": ["mcp", "--db", "./raven.db"]
    }
  }
}

Or configure globally in VS Code settings (settings.json):

{
  "github.copilot.chat.mcp.servers": {
    "ravenrag": {
      "command": "ravenrag",
      "args": ["mcp", "--db", "/path/to/raven.db"]
    }
  }
}

After adding, Copilot Chat will automatically discover the tools. You can invoke them with @ravenrag or the agent will use them when relevant.

Docker (any client)¶

If you installed via Docker, use this command format:

{
  "mcpServers": {
    "ravenrag": {
      "command": "docker",
      "args": [
        "run", "--rm", "-i",
        "-v", "raven-data:/data",
        "ghcr.io/egkristi/ravenrustrag:latest",
        "mcp", "--db", "/data/raven.db"
      ]
    }
  }
}

Available Tools¶

search¶

Query the index with semantic search.

Parameters: - query (string, required): Search query text - top_k (integer, optional): Number of results (default: 5)

get_prompt¶

Search and format results as an LLM-ready prompt with citations.

Parameters: - query (string, required): Query text - top_k (integer, optional): Number of context chunks (default: 3)

collection_info¶

Get index statistics (document count, chunk count, database size).

No parameters required.

index_documents¶

Add documents to the index.

Parameters: - documents (array, required): Array of { content, metadata } objects

Protocol Details¶

The MCP server implements: - initialize / initialized handshake - tools/list — returns available tools (respects --filter) - tools/call — executes a tool - resources/list — lists browseable resources - resources/read — reads a resource by URI - prompts/list — lists available prompt templates - prompts/get — renders a prompt template with arguments - Proper error responses with JSON-RPC error codes

MCP Protocol Version: 2024-11-05

Resources¶

Resources expose index metadata as browseable URIs.

Prompts¶

Prompt templates that AI assistants can invoke:

rag_answer¶

Generate an answer using retrieved context.

Arguments: - query (string, required): The question to answer - top_k (string, optional): Number of context chunks (default: "3")

summarize_index¶

Summarize the contents of the document index.

No arguments required.

Input Validation¶

Input validation: - Queries are limited to 10,000 characters - top_k must be between 1 and 100 - Batch indexing is limited to 100 documents per call