Fast Search MCP Server

AI-powered semantic code search for your IDE using the Model Context Protocol (MCP). Get intelligent, context-aware search results powered by Cerebras AI and ripgrep.

Features

• 🚀 Blazing Fast: Leverages ripgrep for lightning-fast file searching
• 🧠 AI-Powered: Uses Cerebras LLM to understand semantic queries
• 🔍 Smart Search: Finds relevant code even with natural language queries
• 🌐 HTTP/2 Support: End-to-end HTTP/2 for optimal performance
• 🔌 MCP Compatible: Works with any MCP-compatible IDE (Kiro, Claude Desktop, etc.)

Installation

Global Installation (Recommended)

npm install -g @cheetah-ai/fast-search-mcp

Local Installation

npm install @cheetah-ai/fast-search-mcp

Configuration

1. Get Your API Key

You need a Cerebras API key from the proxy service. Contact the administrator or use the public endpoint.

2. Configure MCP

Add to your MCP configuration file:

For Kiro IDE (.kiro/settings/mcp.json):

{
  "mcpServers": {
    "fast-search": {
      "command": "npx",
      "args": ["@cheetah-ai/fast-search-mcp"],
      "env": {
        "CEREBRAS_API_KEY": "your-api-key-here",
        "CEREBRAS_API_URL": "https://cerebras-proxy-rrrczy5zva-uc.a.run.app/api",
        "CEREBRAS_MODEL": "llama-3.3-70b"
      },
      "disabled": false,
      "autoApprove": ["semantic_code_search"]
    }
  }
}

For Claude Desktop (~/Library/Application Support/Claude/claude_desktop_config.json on macOS):

{
  "mcpServers": {
    "fast-search": {
      "command": "npx",
      "args": ["@cheetah-ai/fast-search-mcp"],
      "env": {
        "CEREBRAS_API_KEY": "your-api-key-here",
        "CEREBRAS_API_URL": "https://cerebras-proxy-rrrczy5zva-uc.a.run.app/api",
        "CEREBRAS_MODEL": "llama-3.3-70b"
      }
    }
  }
}

3. Environment Variables

| Variable | Description | Default | |----------|-------------|---------| | CEREBRAS_API_KEY | Your API key (required) | - | | CEREBRAS_API_URL | Proxy endpoint URL | https://cerebras-proxy-rrrczy5zva-uc.a.run.app/api | | CEREBRAS_MODEL | Model to use | llama-3.3-70b | | DEFAULT_MAX_RESULTS | Max results per search | 100 | | MAX_PARALLEL_PROCESSES | Concurrent searches | 5 |

Usage

Once configured, use the semantic search tool in your IDE:

Example Queries

"Find authentication logic"
"Where is the user login handled?"
"Show me database connection code"
"Find React components that use hooks"
"Locate error handling middleware"

MCP Tool

The server exposes one tool:

semantic_code_search

• Input:
  • query (string): Natural language search query
  • path (string): Absolute path to codebase
• Output: List of relevant file paths with scores

How It Works

1. Query Processing: Your natural language query is sent to Cerebras AI
2. Pattern Generation: AI generates multiple search patterns and keywords
3. Fast Search: Ripgrep executes parallel searches across your codebase
4. Smart Ranking: Results are scored and ranked by relevance
5. Return Results: Top matches are returned to your IDE

Requirements

• Node.js >= 18.0.0
• ripgrep installed and available in PATH
  • macOS: brew install ripgrep
  • Ubuntu/Debian: apt install ripgrep
  • Windows: choco install ripgrep or download from releases

Development

Run Locally

git clone https://github.com/cheetah-ai/fast-search-mcp.git
cd fast-search-mcp
npm install
node src/mcp-server.js

Test

npm test

Architecture

┌─────────────┐
│   IDE/MCP   │
└──────┬──────┘
       │ MCP Protocol
       ▼
┌─────────────────┐
│  MCP Server     │
│  (This Package) │
└──────┬──────────┘
       │ HTTP/2
       ▼
┌─────────────────┐
│ Cerebras Proxy  │
│  (Cloud Run)    │
└──────┬──────────┘
       │
       ▼
┌─────────────────┐
│  Cerebras API   │
└─────────────────┘

Troubleshooting

"ripgrep not found"

Install ripgrep: https://github.com/BurntSushi/ripgrep#installation

"API key invalid"

Check your CEREBRAS_API_KEY in the MCP configuration.

"Connection timeout"

Verify the CEREBRAS_API_URL is accessible and correct.

MCP server not starting

Check Node.js version: node --version (must be >= 18)

License

MIT

Support

• Issues: https://github.com/cheetah-ai/fast-search-mcp/issues
• Documentation: https://github.com/cheetah-ai/fast-search-mcp

Credits

Built with:

• Cerebras AI - Ultra-fast LLM inference
• ripgrep - Fast text search
• Model Context Protocol - MCP standard