Custom Extensions & Automation

Goose's MCP-native architecture means you can build your own MCP servers to give Goose access to any tool, API, or system. A custom MCP server is a simple process that exposes tools through the MCP protocol. This lesson covers building a custom extension and automating workflows with Goose.

What you'll learn

• Custom MCP servers let Goose access any tool, API, or system you control
• MCP servers can be written in any language (Python, Node, Go, Rust)
• Automate workflows by chaining MCP servers: database query + report generation + Slack notification

Build it

Create a new Python file and instantiate an MCP server to expose your custom deployment tools. Decorate your functions with the @server.tool() decorator to make them available for Goose to invoke.

# custom_mcp_server.py
from mcp.server import Server, stdio_server

server = Server("my-tools")

@server.tool()
def deploy_service(service_name: str, environment: str = "staging"):
    """Deploy a service to the specified environment."""
    # Your deployment logic here
    return f"Deployed {service_name} to {environment}"

@server.tool()
def check_service_health(service_name: str):
    """Check the health of a deployed service."""
    # Your health check logic here
    return f"{service_name}: healthy"

stdio_server.run(server)

Register the MCP server with your local Goose instance. Once registered, the custom tools become accessible in any new session.

# Register with Goose
goose mcp add my-tools -- python custom_mcp_server.py

What goes wrong

Mistake	How you notice it	The fix
MCP server crashes silently	Tools stop working mid-session	Add logging to your MCP server. Check Goose logs for connection errors
Tool has side effects during testing	Real services get deployed during development	Use a test environment or mock the deployment logic during development

Next: Capstone: Multi-Tool Dev Workflow