Overview

Defines how to build MCP scripts under the scripts/ folder with strict CLI parameterization, docstring usage, and a standardized tool invocation pattern. It codifies the DOs and DON'Ts (use argparse, include a usage docstring, use call_mcp_tool, avoid hardcoding), and ensures outputs are printed to stdout for Claude to process.

How This Skill Works

Scripts must parse all parameters with argparse, include a top-of-file usage docstring, and invoke MCP tools using call_mcp_tool("server__tool", params). Errors are handled gracefully with informative messages, and results are printed to stdout so Claude can process them.

When to Use It

• You need to run server tools from a script in scripts/ with CLI arguments
• You want to avoid hardcoding parameters and rely on CLI inputs
• You must include a top-of-file usage docstring for clarity
• You want to invoke tools via the call_mcp_tool("server__tool", params) pattern
• You require clean stdout output and graceful error messages for Claude

Quick Start

1. Step 1: Create scripts/<script>.py with a CLI via argparse
2. Step 2: Add a top-of-file usage docstring and use call_mcp_tool("server__tool", params)
3. Step 3: Run uv run python -m runtime.harness scripts/<script>.py --help

Best Practices

• Use argparse for all parameters (no hard-coded values)
• Include a top-of-file usage docstring in every script
• Invoke MCP tools with call_mcp_tool("server__tool", params)
• Avoid editing parameters in the script; rely on CLI args
• Print results to stdout and handle errors with informative messages

Example Use Cases

• morph__warpgrep_codebase_search: scripted tool call with CLI args
• ast-grep__ast_grep: parameterized script invocation
• perplexity__perplexity_ask: CLI-driven questions
• script using runtime.mcp_client instead of importing servers/ directly
• script that includes a top-of-file usage docstring and prints stdout

Frequently Asked Questions