Model Context Protocol (MCP)

The Genkit MCP plugin provides integration between Genkit and the Model Context Protocol (MCP). The plugin allows you to:

• Consume MCP tools as a client using create_mcp_client or create_mcp_host
• Expose Genkit tools as an MCP server using create_mcp_server

Installation

pip install genkit-plugin-mcp

MCP Client

Connect to MCP servers to use their tools:

import asyncio

from genkit import Genkit
from genkit.plugins.mcp import McpServerConfig, create_mcp_client
from genkit.plugins.google_genai import GoogleAI

async def main():
    # Create a client for an MCP server
    fs_client = create_mcp_client(
        name='filesystem',
        config=McpServerConfig(
            command='npx',
            args=['-y', '@modelcontextprotocol/server-filesystem', '.'],
        ),
    )

    ai = Genkit(
        plugins=[GoogleAI(), fs_client],
        model='googleai/gemini-2.5-flash',
    )

    # Connect to the server
    await fs_client.connect()

    # List available tools
    tools = await fs_client.list_tools()
    for tool in tools:
        print(f'- {tool.name}: {tool.description}')

    # Use tools in generation
    response = await ai.generate(
        prompt='List all Python files in the current directory',
        tools=[tool.name for tool in tools],
    )
    print(response.text)

    await fs_client.close()

asyncio.run(main())

MCP Host

Connect to multiple MCP servers at once:

from genkit.plugins.mcp import McpServerConfig, create_mcp_host

# Create a host managing multiple servers
host = create_mcp_host({
    'filesystem': McpServerConfig(
        command='npx',
        args=['-y', '@modelcontextprotocol/server-filesystem', '.'],
    ),
    'memory': McpServerConfig(
        command='npx',
        args=['-y', '@modelcontextprotocol/server-memory'],
    ),
})

# Start all servers
await host.start()

# Register tools from all connected servers
await host.register_tools(ai)

# Close all connections
await host.close()

MCP Server

Expose your Genkit tools as an MCP server:

import asyncio
from pydantic import BaseModel, Field
from genkit import Genkit
from genkit.plugins.mcp import McpServerOptions, create_mcp_server

class AddInput(BaseModel):
    a: int = Field(description='First number')
    b: int = Field(description='Second number')

ai = Genkit()

@ai.tool(name='add', description='Add two numbers together')
def add(input: AddInput) -> int:
    return input.a + input.b

# Create and start the MCP server
server = create_mcp_server(
    ai,
    McpServerOptions(name='my-genkit-server', version='1.0.0'),
)

# Run the server (stdio transport)
asyncio.run(server.start())

Configuration Options

McpServerConfig

McpServerConfig(
    command='npx',           # Command to run
    args=['server-name'],    # Command arguments
    url='http://...',        # Alternative: HTTP endpoint
    disabled=False,          # Disable this server
)

McpServerOptions

McpServerOptions(
    name='my-server',        # Server name
    version='1.0.0',         # Server version
)

Using MCP Tools with AI

from genkit import Genkit
from genkit.plugins.google_genai import GoogleAI
from genkit.plugins.mcp import create_mcp_client, McpServerConfig

client = create_mcp_client(
    name='tools-server',
    config=McpServerConfig(command='my-mcp-server'),
)

ai = Genkit(plugins=[GoogleAI(), client])

await client.connect()

# Get tools and use them in generation
tools = await client.list_tools()
response = await ai.generate(
    prompt='Use the available tools to help me',
    tools=[tool.name for tool in tools],
)

await client.close()