regenerator2000: Skill for Claude Code

Add MCP Tool Workflow

Use this workflow when adding a new tool to the crates/regenerator2000-core/src/mcp/handler.rs server.

1. Plan the Tool

Before writing code, confirm with the user:

• Tool Name (e.g., r2000_get_memory)
• Arguments (e.g., address: u16, length: u16)
• Return Type (e.g., Vec<u8>)
• Description

2. Define the Tool in list_tools

In crates/regenerator2000-core/src/mcp/handler.rs:

• Use define_tool! macro or manually add the JSON definition inside the list_tools function.
• Ensure arguments follow the JSON schema format.

Example:

json!({
    "name": "r2000_my_tool",
    "description": "Description of what the tool does.",
    "inputSchema": {
        "type": "object",
        "properties": {
            "arg1": { "type": "string", "description": "..." }
        },
        "required": ["arg1"]
    }
})

3. Implement the Handler Logic

In crates/regenerator2000-core/src/mcp/handler.rs:

• Locate handle_tool_call_internal.
• Add a new match arm for your tool name.
• Call a dedicated implementation function (create one if it doesn't exist).

Example:

"r2000_my_tool" => {
    let arg1 = args["arg1"].as_str().ok_or(McpError::InvalidParams("Missing arg1".to_string()))?;
    let result = my_tool_impl(app_state, arg1)?;
    Ok(json!({ "content": [{ "type": "text", "text": result }] }))
}

4. Create the Implementation Function

In crates/regenerator2000-core/src/mcp/handler.rs (or a sub-module):

• Create a function named [tool_name]_impl.
• Accept &mut AppState (or &AppState if read-only).
• Return Result<Value, McpError> or a specific type.

5. Add Verification Test

In tests/verify_mcp.py:

• Create a new function test_[tool_name](client).
• Use client.rpc("tools/call", { ... }).
• Verify the result ("PASS" or "FAIL").
• Add the function call to the if __name__ == "__main__": block.

6. Verify Correctness

• Run the verify-mcp skill:

  .agent/skills/verify-mcp/scripts/verify.sh
  

• Fix any compilation errors or test failures.