Multi-Tool Pipeline Template

Reference implementation showing how to chain multiple MCP tools in a script.

When to Use

• As a template when creating new MCP pipeline scripts
• To understand the pattern for tool chaining
• When skill-developer needs to create a new pipeline

The Pattern

async def main():
    from runtime.mcp_client import call_mcp_tool

    # Step 1: First tool
    result1 = await call_mcp_tool("server1__tool1", {"param": "value"})

    # Step 2: Use result in next tool
    result2 = await call_mcp_tool("server2__tool2", {"input": result1})

    # Step 3: Combine/process
    return {"combined": result1, "processed": result2}

Example Implementation

See the reference script:

cat $CLAUDE_PROJECT_DIR/scripts/multi_tool_pipeline.py

Run it:

uv run python -m runtime.harness scripts/multi_tool_pipeline.py \
    --repo-path "." \
    --max-commits 5

Key Elements

1. CLI Arguments - Use argparse for parameters
2. Sequential Calls - await each tool before the next
3. Error Handling - try/except around the pipeline
4. Progress Output - print status for visibility
5. Structured Return - return combined results

Creating Your Own Pipeline

1. Copy scripts/multi_tool_pipeline.py as a starting point
2. Replace the tool calls with your MCP servers/tools
3. Adjust CLI arguments for your use case
4. Use /skill-developer to wrap it in a skill

MCP Tool Naming

Tools are named serverName__toolName (double underscore):

await call_mcp_tool("git__git_status", {...})
await call_mcp_tool("firecrawl__firecrawl_scrape", {...})
await call_mcp_tool("perplexity__perplexity_ask", {...})