Trace MCP with BotGauge

Records the MCP client and server.


Extra	`[mcp]`
Verified framework versions	mcp 1.24.0 and 2.3.0
Instrumentor installed by the extra	`openinference-instrumentation-mcp 2.0.13`
Zero-code run	Traced

Also verified: mcp 1.24.0 with instrumentor 2.0.10. The extra no longer installs that older instrumentor. See supported frameworks for how verification works.

In code

Call tracing.init(), attach the framework instrumentor to the provider it returns, and wrap the work in tracing.turn() and tracing.agent().

In this sample an LLM, through the OpenAI SDK, calls a tool on an MCP server running in the same process. The [mcp] extra traces the MCP client and server, and [openai] traces the LLM calls.

Install

terminal

pip install "botgauge[mcp,openai]" mcp==2.3.0 openai==3.35.1

Add the code

app.py

import json
import os

import anyio
from botgauge import tracing
from mcp import Client
from mcp.server.mcpserver import MCPServer
from openai import AsyncOpenAI
from openinference.instrumentation.mcp import MCPInstrumentor
from openinference.instrumentation.openai import OpenAIInstrumentor

provider = tracing.init()
MCPInstrumentor().instrument(tracer_provider=provider)
OpenAIInstrumentor().instrument(tracer_provider=provider)

server = MCPServer("accounts")
llm = AsyncOpenAI()

@server.tool()
def lookup_account(account_id: str) -> str:
    """Look up an account by its id."""
    with tracing.get_tracer().start_as_current_span("lookup_account"):
        return f"account {account_id} is active"

async def ask(client: Client, question: str) -> str:
    listed = await client.list_tools()
    tools = [{"type": "function", "function": {"name": t.name, "description": t.description,
                                               "parameters": t.input_schema}} for t in listed.tools]
    messages = [{"role": "user", "content": question}]
    first = await llm.chat.completions.create(model=os.environ["OPENAI_MODEL"], messages=messages, tools=tools)
    reply = first.choices[0].message
    messages.append(reply)
    for call in reply.tool_calls or []:
        result = await client.call_tool(call.function.name, json.loads(call.function.arguments))
        messages.append({"role": "tool", "tool_call_id": call.id, "content": result.content[0].text})
    final = await llm.chat.completions.create(model=os.environ["OPENAI_MODEL"], messages=messages)
    return final.choices[0].message.content

async def main():
    async with Client(server) as client:
        return await ask(client, "Is account ACC001 active? Check with the tool.")

with tracing.turn(session_id="ses_mcp"), tracing.agent("mcp-agent"):
    answer = anyio.run(main)
print(answer)

tracing.shutdown()

Run

terminal

python app.py

Needs BOTGAUGE_API_KEY, OPENAI_API_KEY, OPENAI_MODEL. Set OPENAI_BASE_URL too if your LLM isn’t api.openai.com.

Zero-code

Leave your app unchanged and start it with botgauge instrument. The sample still wraps the work in tracing.turn() and tracing.agent() so spans carry session and agent ids.

In this sample an LLM, through the OpenAI SDK, calls a tool on an MCP server running in the same process. The [mcp] extra traces the MCP client and server, and [openai] traces the LLM calls.

Install

terminal

pip install "botgauge[mcp,openai]" mcp==2.3.0 openai==3.35.1

Add the code

app.py

import json
import os

import anyio
from botgauge import tracing
from mcp import Client
from mcp.server.mcpserver import MCPServer
from openai import AsyncOpenAI

server = MCPServer("accounts")
llm = AsyncOpenAI()

@server.tool()
def lookup_account(account_id: str) -> str:
    """Look up an account by its id."""
    with tracing.get_tracer().start_as_current_span("lookup_account"):
        return f"account {account_id} is active"

async def ask(client: Client, question: str) -> str:
    listed = await client.list_tools()
    tools = [{"type": "function", "function": {"name": t.name, "description": t.description,
                                               "parameters": t.input_schema}} for t in listed.tools]
    messages = [{"role": "user", "content": question}]
    first = await llm.chat.completions.create(model=os.environ["OPENAI_MODEL"], messages=messages, tools=tools)
    reply = first.choices[0].message
    messages.append(reply)
    for call in reply.tool_calls or []:
        result = await client.call_tool(call.function.name, json.loads(call.function.arguments))
        messages.append({"role": "tool", "tool_call_id": call.id, "content": result.content[0].text})
    final = await llm.chat.completions.create(model=os.environ["OPENAI_MODEL"], messages=messages)
    return final.choices[0].message.content

async def main():
    async with Client(server) as client:
        return await ask(client, "Is account ACC001 active? Check with the tool.")

with tracing.turn(session_id="ses_mcp"), tracing.agent("mcp-agent"):
    answer = anyio.run(main)
print(answer)

Run

terminal

botgauge instrument python app.py

Needs BOTGAUGE_API_KEY, OPENAI_API_KEY, OPENAI_MODEL. Set OPENAI_BASE_URL too if your LLM isn’t api.openai.com.

View MCP spans live

Watch your MCP client, server, and LLM tool calls appear in BotGauge.

Try BotGauge →

See versions not verified if you run a different version, capture and redaction to control exported content, and troubleshooting if spans are missing.

Frameworks

More

In code

Zero-code

​In code

​Zero-code

​Related

In code

Zero-code

Related