Skip to content

sample-tool.md

Latest commit

 

History

History
51 lines (33 loc) · 1.15 KB

sample-tool.md

File metadata and controls

51 lines (33 loc) · 1.15 KB

Sample Tool: speak_<animal>

The scaffolded plugin includes a single MCP tool that demonstrates the self-installing pattern working end-to-end.

What it generates

For scaffold_plugin("alligator", "After while, crocodile"):

"""MCP server for alligator-speak."""

from mcp.server.fastmcp import FastMCP

mcp = FastMCP("alligator-speak")


@mcp.tool()
def speak_alligator() -> str:
    """Say something like a alligator."""
    return "After while, crocodile"


if __name__ == "__main__":
    mcp.run(transport="stdio")

What this proves

When this tool responds correctly, you know:

  • The SessionStart hook installed mcp into the plugin's isolated site-packages
  • PYTHONPATH is set correctly in .mcp.json
  • The MCP server started via python3 and connected over stdio
  • Claude discovered and invoked the tool by matching user intent

Expected output

> speak alligator

🐊 "After while, crocodile!"

Next steps

Rename speak_alligator to your real tool, update server.py with your actual logic, and add dependencies to requirements.txt. The plugin infrastructure (hooks, .mcp.json, plugin.json) stays the same.