instrument_mcp docs (#1412)

Author: alexmojaki

docs/images/logfire-screenshot-mcp.png

@@ -0,0 +1,61 @@
+---
+integration: logfire
+---
+# Model Context Protocol (MCP)
+
+
+**Logfire** supports instrumenting the [MCP Python SDK](https://github.com/modelcontextprotocol/python-sdk) with the [`logfire.instrument_mcp()`][logfire.Logfire.instrument_mcp] method. This works on both the client and server side. If possible, calling this in both the client and server processes is recommended for nice distributed traces.
+
+Below is a simple example. For the client, we use [Pydantic AI](https://ai.pydantic.dev/mcp/client/) (though any MCP client will work) and OpenAI. To use a different LLM provider instead of OpenAI, replace `openai:gpt-4o` in the client script with a different model name supported by Pydantic AI.
+
+First, install the required dependencies:
+
+```bash
+pip install mcp 'pydantic-ai-slim[openai]'
+```
+
+Next, run the server script below:
+
+```python title="server.py"
+from mcp.server.fastmcp import FastMCP
+
+import logfire
+
+logfire.configure(service_name='server')
+logfire.instrument_mcp()
+
+app = FastMCP()
+
+
+@app.tool()
+def add(a: int, b: int) -> int:
+    logfire.info(f'Calculating {a} + {b}')
+    return a + b
+
+
+app.run(transport='streamable-http')
+```
+
+Then run this client script in another terminal:
+
+```python title="agent.py"
+from pydantic_ai import Agent
+from pydantic_ai.mcp import MCPServerStreamableHTTP
+
+import logfire
+
+logfire.configure(service_name='agent')
+logfire.instrument_pydantic_ai()  # (1)!
+logfire.instrument_mcp()
+
+server = MCPServerStreamableHTTP('http://localhost:8000/mcp')
+agent = Agent('openai:gpt-4o', toolsets=[server])
+result = agent.run_sync('What is 7 plus 5?')
+print(result.output)
+```
+
+1. Instrumenting Pydantic AI is optional, but adds more context to the trace.
+
+You should see a trace like this in Logfire:
+
+![Logfire MCP Trace](../../images/logfire-screenshot-mcp.png)

docs/integrations/llms/mcp.md

@@ -913,10 +913,14 @@ def _warn_if_not_initialized_for_instrumentation(self):
         self.config.warn_if_not_initialized('Instrumentation will have no effect')
 
     def instrument_mcp(self, *, propagate_otel_context: bool = True) -> None:
-        """Instrument [MCP](https://modelcontextprotocol.io/) requests such as tool calls.
+        """Instrument the [MCP Python SDK](https://github.com/modelcontextprotocol/python-sdk).
+
+        Instruments both the client and server side. If possible, calling this in both the client and server
+        processes is recommended for nice distributed traces.
 
         Args:
-            propagate_otel_context: Whether to enable propagation of the OpenTelemetry context.
+            propagate_otel_context: Whether to enable propagation of the OpenTelemetry context
+                for distributed tracing.
                 Set to False to prevent setting extra fields like `traceparent` on the metadata of requests.
         """
         from .integrations.mcp import instrument_mcp

logfire/_internal/main.py

@@ -117,6 +117,7 @@ nav:
           - Google Gen AI: integrations/llms/google-genai.md
           - Anthropic: integrations/llms/anthropic.md
           - LangChain: integrations/llms/langchain.md
+          - MCP: integrations/llms/mcp.md
           - LLamaIndex: integrations/llms/llamaindex.md
           - Mirascope: integrations/llms/mirascope.md
           - LiteLLM: integrations/llms/litellm.md