Hey, I have implemented a connection of my agent with an mcp server. I just added the mcp_servers=[mcp.MCPServerHTTP(url="...", timeout=5.0)] Is there a way to log call and responses to and from mcp servers? I have no visibility what the agent is send to the server nor what is receiving

Thanks for reaching out, Roman. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.

Seems like my logger is broker. This error was introduced by installing

uv pip install "livekit-agents[mcp]"

211134.373 DEBUG livekit.agents http_session(): creating a new httpclient ctx {"room": "voice_assistant_room_5127", "agent": "assistant", "lang": "es", "pid": 14509, "job_id": "AJ_NTgrbbVqDBaa", "room_id": "RM_LkFTNZ54hkQ3"} VAAPI is supported. Exception in thread ipc_log_listener: Traceback (most recent call last): File "/home/roman/.pyenv/versions/3.12.12/lib/python3.12/threading.py", line 1075, in _bootstrap_inner self.run() File "/home/roman/.pyenv/versions/3.12.12/lib/python3.12/threading.py", line 1012, in run self._target(*self._args, **self._kwargs) File "/home/roman/Documents/projects/wil-livekit-worker/.venv/lib/python3.12/site-packages/livekit/agents/ipc/log_queue.py", line 53, in _monitor record = pickle.loads(data) ^^^^^^^^^^^^^^^^^^ TypeError: HTTPStatusError.__init__() missing 2 required keyword-only arguments: 'request' and 'response' ^C 211217.954 INFO livekit.agents shutting down worker {"id": "AW_L5NQbbQJjvm7"} 211218.220 INFO livekit.agents process exiting {"reason": "", "pid": 14509, "job_id": "AJ_NTgrbbVqDBaa", "room_id": "RM_LkFTNZ54hkQ3"}

Thanks — you're reporting the IPC logging crash after installing

livekit-agents[mcp]

and asking how to restore logging and capture MCP requests/responses. You're right: that TypeError indicates a version/serialization mismatch for

HTTPStatusError

. What to try next • Pin httpx to a compatible version:

uv pip install "httpx==0.25.2"

If needed also pin httpcore:

uv pip install "httpcore==1.0.5"

Then restart your agent process. • Avoid dev mode while debugging (dev uses IPC/pickle which triggers the error). Use:

uv run agent.py start

• Fully restart your environment to clear any pickled state or lingering processes: ◦ Kill agent processes, stop hot-reload, then start fresh. • If you need immediate visibility regardless of the IPC issue, add a small wrapper around MCP calls to log requests and responses (example):

import logging
  logger = logging.getLogger("mcp_debug")

  async def _wrap_call(method, *args, **kwargs):
      logger.debug("MCP CALL %s %s %s", method, args, kwargs)
      try:
          resp = await method(*args, **kwargs)
          logger.debug("MCP RESP %s", resp)
          return resp
      except Exception as e:
          logger.exception("MCP ERROR")
          raise
  # attach wrapper to the MCPServerHTTP methods you use (e.g. list_tools, call_tool)

If this doesn't fix it, please share these outputs: •

python --version

•

uv --version

•

uv pip list | grep -E "(httpx|httpcore|livekit)"

• How you run the agent (CLI command:

dev

vs

start

) I’ll give the exact next steps based on those details.

Still I wasn't able to make logs work. This are the versions I'm using:

roman@roman-pc:~/Documents/projects/wil-livekit-worker$ python --version
Python 3.12.12
roman@roman-pc:~/Documents/projects/wil-livekit-worker$ uv --version
uv 0.9.17
roman@roman-pc:~/Documents/projects/wil-livekit-worker$ uv pip list | grep -E "(httpx|httpcore|livekit)"
agent-starter-python                     1.0.0           /home/roman/Documents/projects/wil-livekit-worker
httpcore                                 1.0.9
httpx                                    0.28.1
httpx-sse                                0.4.3
livekit                                  1.0.22
livekit-agents                           1.3.6
livekit-api                              1.1.0
livekit-blingfire                        1.1.0
livekit-plugins-inworld                  1.3.6
livekit-plugins-noise-cancellation       0.2.5
livekit-plugins-silero                   1.3.6
livekit-plugins-speechify                1.3.6
livekit-plugins-speechmatics             1.3.6
livekit-plugins-turn-detector            1.3.6
livekit-protocol                         1.1.1

Is there any recommended versions for this?

Thanks — I see your environment and versions, and you’re still not getting logs. The

httpx 0.28.1

/

httpcore 1.0.9

combo is incompatible with the agent IPC logging and causes the pickling/HTTPStatusError problem. What to do next 1. Pin compatible versions:

uv pip install "httpx==0.27.2" "httpcore==1.0.7"

1. Restart clean: • Kill all agent processes. • Stop dev/hot-reload if you use it. • Start the agent with

uv run agent.py start

. 1. If you use uv-managed constraints, add these to your project (example

pyproject.toml

snippet):

[tool.uv]
constraints = [
  "httpx==0.27.2",
  "httpcore==1.0.7"
]

then run:

uv lock
uv sync --group dev

1. Optional: If you can, use Python 3.11 for best compatibility (3.12 may surface other issues). If logs still fail after the downgrade and restart, please paste the output of: •

uv pip list | grep -E "(httpx|httpcore|livekit)"

• How you run the agent (exact CLI command:

dev

or

start

) I’ll give the next troubleshooting step based on that. Sources: https://blog.livekit.io/streamline-troubleshooting-with-agent-observability/#/portal/signup| | Streamline Troubleshooting with Agent Observability | https://blog.livekit.io/streamline-troubleshooting-with-agent-observability/#/portal| | Tool definition and use | LiveKit docs | Voice AI quickstart | LiveKit docs

How do I make sure the livekit agent has access to the MCP server that I have connected to it?

You're asking how to make sure your LiveKit agent can reach and use the MCP server you connected. Here are concise, practical checks and actions you can run now. 1) Confirm configuration • Ensure you passed the MCP server into your session:

from livekit.agents import mcp, AgentSession
session = AgentSession(mcp_servers=[mcp.MCPServerHTTP(url="<https://your-mcp-server.com>")])

2) Check network reachability from the agent host • Simple HTTP test:

curl -v <https://your-mcp-server.com/>
# or, for tool listing endpoint:
curl -v -X POST <https://your-mcp-server.com/tools.list>

• If curl fails, check DNS, firewall, proxy, or container network settings. 3) Watch agent startup logs • On startup the agent lists/discovers MCP tools. Look for lines about MCP initialization, tool discovery, or HTTP errors. If you need logging, run with DEBUG once your httpx/httpcore versions are compatible:

uv run agent.py start --log-level=DEBUG

4) Verify tool discovery at runtime • Inspect

session.tools

or the agent’s available tools in code or logs to confirm MCP-provided tools were merged with local tools. If MCP tools appear, connectivity is working. 5) Trigger a known MCP tool call • Ask the agent to call a tool that the MCP server provides (or invoke it directly). Successful invocation and a valid response confirm end-to-end access. 6) Use LiveKit Agent observability (if available) • In LiveKit Cloud, check Agent Insights: trace view for LLM tool call attempts, logs for HTTP errors, and transcripts for tool usage. 7) Authentication & headers • If your MCP requires auth or headers, ensure you provide them in

MCPServerHTTP(url=..., headers={...})

or via the server’s configuration. 8) If you see errors, capture and share: • Agent CLI command you run (exact

uv run ...

) • Agent logs around MCP init (DEBUG if possible) • Output of: ◦

curl -v <https://your-mcp-server.com/>

(or

tools.list

) ◦

uv pip list | grep -E "(httpx|httpcore|livekit)"

If you want, tell me how you currently run the agent and the MCP URL (or a curl output). I’ll suggest the next targeted steps. Sources: Tool definition and use | LiveKit docs | MCP Agent | LiveKit docs | Agents (Python) | livekit/livekit_composite | DeepWiki | https://blog.livekit.io/streamline-troubleshooting-with-agent-observability/#/portal/signup|