Skip to main content

Using Saiki as an MCP Server

Saiki agents can act as Model Context Protocol (MCP) server, enabling external tools like Cursor/Claude Desktop or any MCP client to connect and interact with your Saiki agent.

This means you can even connect one Saiki agent to another Saiki agent!

The default Saiki agent has tools to access files and browse the web, but you can configure this too by changing the config file!

Check out our Configuration guide

Local MCP Server Guide​

Setup in Cursor​

  1. Create or edit your .cursor/mcp.json file:

Use the default Saiki configuration

{
  "mcpServers": {
    "saiki": {
      "command": "npx",
      "args": ["-y", "@truffle-ai/saiki", "--mode", "mcp"],
      "env": {
        "OPENAI_API_KEY": "your_openai_api_key"
      }
    }
  }
}

Using a custom Saiki configuration: Note: if you use a different LLM in your config file, you will need to pass the appropriate environment variable for that provider.

{
  "mcpServers": {
    "saiki": {
      "command": "npx",
      "args": ["-y", "@truffle-ai/saiki", "--mode", "mcp", "--agent", "path/to/your/agent.yml"],
      "env": {
        "OPENAI_API_KEY": "your_openai_api_key"
      }
    }
  }
}
  1. Restart Cursor

Using Saiki in Cursor​

Available Tools in Cursor:

  • chat_with_agent: Interact with Saiki AI agent

Available Saiki tools:

By default, Saiki CLI loads an AI agent that has tools to:

  • browse the web
  • search files on your local system

But you can customize the tools by using a custom Saiki agent configuration file. Check out our Configuration guide.

Example Usage in Cursor:

  1. Refactor a function:

    Ask Saiki agent to help me refactor this function to be more efficient

  2. Get file analysis:

    Ask Saiki agent to analyze the architecture of this project

  3. Browse the web:

    Ask Saiki agent to search the web for soccer shoes under $100

  4. Any custom functionality: You can configure your Saiki agent to have any other custom functionality by setting up your own config file and using it here. Check out our Configuration guide

Remote MCP Server Setup​

Step 1: Start Saiki in Server Mode​

# If installed globally
saiki --mode server

# Or via npx
npx @truffle-ai/saiki --mode server

Options:

# Custom port using environment variable
API_PORT=8080 saiki --mode server
# Or via npx
API_PORT=8080 npx @truffle-ai/saiki --mode server

# Custom port for network access
API_PORT=3001 saiki --mode server
# Or via npx
API_PORT=3001 npx @truffle-ai/saiki --mode server

# Enable debug logging
saiki --mode server --debug
# Or via npx
npx @truffle-ai/saiki --mode server --debug

Step 2: Configure the Connection URL​

HTTP MCP Endpoint:

http://localhost:3001/mcp

For network access:

http://YOUR_SERVER_IP:3001/mcp

Remote Server in Cursor (WIP)​

Cursor/Claude desktop don't support streamable http yet

Troubleshooting​

Cursor not detecting MCP server:

  • Verify .cursor/mcp.json syntax is correct
  • Restart Cursor after configuration changes
  • Ensure @truffle-ai/saiki is installed and accessible
  • Verify environment variables are set correctly

Debug mode:

# If installed globally
saiki --mode mcp --debug

# Or via npx
npx @truffle-ai/saiki --mode mcp --debug