Grok MCP

Implementation Reference

• src/server.py:341-363 (handler)

  The code_executor tool handler function. It accepts a prompt, optional model (default 'grok-4-1-fast'), and optional max_turns parameter. Creates an XAI client, configures it with code_execution tool, sends the prompt, and returns the response content along with any code execution outputs.

  async def code_executor(
      prompt: str,
      model: str = "grok-4-1-fast",
      max_turns: Optional[int] = None
  ):
      client = Client(api_key=XAI_API_KEY)
      
      chat_params = {"model": model, "tools": [code_execution()], "include": ["code_execution_call_output"]}
      if max_turns:
          chat_params["max_turns"] = max_turns
      
      chat = client.chat.create(**chat_params)
      chat.append(user(prompt))
      response = chat.sample()
      
      client.close()
  
      result = [response.content]
      if response.tool_outputs:
          result.append("\n\n**Code Output(s):**")
          for output in response.tool_outputs:
              result.append(f"```\n{output.message.content}\n```")
      return "\n".join(result)

• src/server.py:340-340 (registration)

  The @mcp.tool() decorator registers the code_executor function as an MCP tool. The mcp instance is a FastMCP server created at line 11.

  @mcp.tool()

• src/server.py:8-8 (helper)

  Import of code_execution tool from xai_sdk.tools, which is used in the code_executor handler to enable code execution capabilities.

  from xai_sdk.tools import web_search as xai_web_search, x_search as xai_x_search, code_execution

• src/server.py:6-6 (helper)

  Import of Client from xai_sdk, used to create the API client instance in the code_executor handler.

  from xai_sdk import Client