MCP Server Overview

The laurus-mcp crate provides a Model Context Protocol (MCP) server for the Laurus search engine. It acts as a gRPC client to a running laurus-server instance, enabling AI assistants such as Claude to index documents and perform searches through the standard MCP stdio transport.

Features

• MCP stdio transport — Runs as a subprocess; communicates with the AI client via stdin/stdout
• gRPC client — Proxies all tool calls to a running laurus-server instance
• All laurus search modes — Lexical (BM25), vector (HNSW/Flat/IVF), and hybrid search
• Dynamic connection — Connect to any laurus-server endpoint via the connect tool
• Document lifecycle — Add, update, delete, and retrieve documents through MCP tools

Architecture

graph LR
    subgraph "laurus-mcp"
        MCP["MCP Server\n(stdio)"]
    end

    AI["AI Client\n(Claude, etc.)"] -->|"stdio (JSON-RPC)"| MCP
    MCP -->|"gRPC"| SRV["laurus-server\n(always running)"]
    SRV --> Disk["Index on Disk"]

The MCP server runs as a child process launched by the AI client. It proxies all tool calls to a laurus-server instance via gRPC. The laurus-server must be started separately before (or after) the MCP server.

Quick Start

# Step 1: Start the laurus-server
laurus serve --grpc-port 50051

# Step 2: Configure Claude Code and start the MCP server
claude mcp add laurus laurus mcp --endpoint http://localhost:50051

Or with a manual configuration:

{
  "mcpServers": {
    "laurus": {
      "command": "laurus",
      "args": ["mcp", "--endpoint", "http://localhost:50051"]
    }
  }
}

Sections

• Getting Started — Installation, configuration, and first steps
• Tools Reference — Full reference for all MCP tools