i'm working with integrating mcp_server in agent, i'm adding following config: to agent to AgentSession, it seems to be not working is there any additional steps to integrate mcp mcp_servers=[ mcp.MCPServerHTTP( url="https://backend.test.com/v3/mcp/esfdsdf?transport=sse", timeout=30, client_session_timeout_seconds=10 ), ]

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

livekit-agents[mcp] is installed here are logs for curl ALPN, offering h2 * ALPN, offering http/1.1 * CAfile: /etc/ssl/certs/ca-certificates.crt * CApath: /etc/ssl/certs * TLSv1.0 (OUT), TLS header, Certificate Status (22): * TLSv1.3 (OUT), TLS handshake, Client hello (1): * TLSv1.2 (IN), TLS header, Certificate Status (22): * TLSv1.3 (IN), TLS handshake, Server hello (2): * TLSv1.2 (IN), TLS header, Finished (20): * TLSv1.2 (IN), TLS header, Supplemental data (23): * TLSv1.3 (IN), TLS handshake, Encrypted Extensions (8): * TLSv1.3 (IN), TLS handshake, Certificate (11): * TLSv1.3 (IN), TLS handshake, CERT verify (15): * TLSv1.3 (IN), TLS handshake, Finished (20): * TLSv1.2 (OUT), TLS header, Finished (20): * TLSv1.3 (OUT), TLS change cipher, Change cipher spec (1): * TLSv1.2 (OUT), TLS header, Supplemental data (23): * TLSv1.3 (OUT), TLS handshake, Finished (20): * SSL connection using TLSv1.3 / TLS_AES_256_GCM_SHA384 * ALPN, server accepted to use h2 * Server certificate: * subject: CN=composio.dev * start date: Nov 10 000553 2025 GMT * expire date: Feb 8 010550 2026 GMT * subjectAltName: host "backend.composio.dev" matched cert's "*.composio.dev" * issuer: C=US; O=Google Trust Services; CN=WE1 * SSL certificate verify ok. * Using HTTP2, server supports multiplexing * Connection state changed (HTTP/2 confirmed) * Copying HTTP/2 data in stream buffer to connection buffer after upgrade: len=0 * TLSv1.2 (OUT), TLS header, Supplemental data (23): * TLSv1.2 (OUT), TLS header, Supplemental data (23): * TLSv1.2 (OUT), TLS header, Supplemental data (23): * Using Stream ID: 1 (easy handle 0x64b149bd89f0) * TLSv1.2 (OUT), TLS header, Supplemental data (23):

GET /v3/mcp/e3edc686-bd73-4363-b869-48dd218799a1?transport=sse HTTP/2

Host: backend.composio.dev

user-agent: curl/7.81.0

accept: /

* TLSv1.2 (IN), TLS header, Supplemental data (23): * TLSv1.3 (IN), TLS handshake, Newsession Ticket (4): * TLSv1.3 (IN), TLS handshake, Newsession Ticket (4): * old SSL session ID is stale, removing * TLSv1.2 (IN), TLS header, Supplemental data (23): * TLSv1.2 (OUT), TLS header, Supplemental data (23): * TLSv1.2 (IN), TLS header, Supplemental data (23): * TLSv1.2 (IN), TLS header, Supplemental data (23): < HTTP/2 307 < date: Tue, 16 Dec 2025 052115 GMT < content-length: 0 < location: https://backend.composio.dev/v3/mcp/e3edc686-bd73-4363-b869-48dd218799a1/sse?transport=sse < server: cloudflare < cf-ray: 9aebbc6fee48a7a6-DEL < cf-cache-status: DYNAMIC < age: 0 < cache-control: public, max-age=0, must-revalidate < strict-transport-security: max-age=63072000 < vary: RSC, Next-Router-State-Tree, Next-Router-Prefetch, Next-Router-Segment-Prefetch < cf-placement: local-DEL < x-matched-path: /v3/mcp/[serverId] < x-vercel-cache: MISS < x-vercel-id: bom1:iad1:v9kjs-1765862474697-fed277665465 < * Connection #0 to host backend.composio.dev left intact

Thanks — I see you have

livekit-agents[mcp]

installed and your curl shows the MCP endpoint returning an HTTP/2 307 redirect to a URL with

/sse

. That helps narrow things down. Next steps to isolate and fix: 1. Confirm agent follows redirects and can reach the final SSE URL ◦ From the agent host run: curl -v "https://backend.composio.dev/v3/mcp/e3edc686-bd73-4363-b869-48dd218799a1/sse?transport=sse" • Paste the response or errors. 1. Share the Agent logs showing MCP connection/tool loading lines ◦ I need the lines around Agent startup and session init that reference MCP, list_tools, timeouts, or errors. Include ~10 lines before/after any MCP-related messages. 1. Check for a 307 handling or SSE-specific issue in the agent logs ◦ If logs show the agent failing to follow the redirect or timing out connecting to the SSE path, that’s likely the problem. 1. If the agent reaches the SSE URL but still no tools, provide the

list_tools

response (JSON) from the MCP server ◦ Run: curl -v "https://…/v3/mcp/<serverId>/list_tools" (or the equivalent endpoint your MCP implements) and paste the JSON. Tell me the curl result for step 1 and paste the MCP-related agent log snippets (step 2). I’ll point to the precise fix once I see those.

getting following in logs ( is it related ) Exception in thread ipc_log_listener: Traceback (most recent call last): File "/home/karandeep/.local/share/uv/python/cpython-3.10.18-linux-x86_64-gnu/lib/python3.10/threading.py", line 1016, in _bootstrap_inner self.run() File "/home/karandeep/.local/share/uv/python/cpython-3.10.18-linux-x86_64-gnu/lib/python3.10/threading.py", line 953, in run self._target(*self._args, **self._kwargs) File "/home/karandeep/Development/livekit-v1/agent-starter-python/.venv/lib/python3.10/site-packages/livekit/agents/ipc/log_queue.py", line 53, in _monitor record = pickle.loads(data) File "/home/karandeep/Development/livekit-v1/agent-starter-python/.venv/lib/python3.10/site-packages/mcp/shared/exceptions.py", line 17, in init super().__init__(error.message) AttributeError: 'str' object has no attribute 'message'

That AttributeError is likely related and will prevent MCP tool info from being processed. It shows an MCP exception object being unpickled where code expects an object with a

message

attribute, but a plain string was found. Please do the following so I can pinpoint the fix: 1. Share package versions ◦ Output of:

pip show livekit-agents mcp

(or

pip list | grep -E "livekit-agents|mcp"

) 1. Search for other MCP/log lines ◦ From your agent logs, paste ~20 lines around any MCP or

list_tools

activity and any tracebacks besides the one you sent. 1. Quick workaround to test if it’s the unpickling path ◦ Restart the agent after enabling DEBUG logs (if not already). If the same error repeats immediately, the MCP path handling is failing early. 1. If you can, try upgrading packages ◦ Run:

uv update livekit-agents[mcp]

(or

pip install -U livekit-agents[mcp] mcp

) and restart the agent, then share the logs again. Once you share the package versions and the additional log lines, I’ll tell you the exact change to make or confirm the package fix. Would you like instructions for collecting DEBUG logs if you don’t already have them?

i am now using agent-starter-python, here's my error when i'm adding mcp_servers livekit.plugins.elevenlabs", "livekit.plugins.openai", "livekit.plugins.deepgram", "av"]} 113510.193 INFO livekit.agents starting inference executor 113510.195 DEV livekit.agents reloading job {"job_id": "AJ_bRujJQ6ihCxy", "agent_name": ""} 113510.844 INFO livekit.agents initializing process {"pid": 510202, "inference": true} 113510.846 INFO livekit.agents initializing process {"pid": 510204} 113511.289 INFO livekit.agents process initialized {"pid": 510204, "elapsed_time": 0.44} 113511.288 DEBUG asyncio Using selector: EpollSelector {"pid": 510204, "job_id": "AJ_bRujJQ6ihCxy", "room_id": "RM_UpQU5TXR6zr6"} 113511.340 DEBUG livekit.agents input stream attached {"participant": null, "source": "SOURCE_UNKNOWN", "accepted_sources": ["SOURCE_MICROPHONE"], "room": "voice_assistant_room_5981", "pid": 510204, "job_id": "AJ_bRujJQ6ihCxy", "room_id": "RM_UpQU5TXR6zr6"} VAAPI is supported. 113513.353 INFO livekit.agents process initialized {"pid": 510202, "inference": true, "elapsed_time": 2.51} 113513.355 INFO livekit.agents HTTP server listening on :41849 113513.611 INFO livekit.agents registered worker {"agent_name": "", "id": "AW_QgLwjKemtYi3", "url": "wss://twil-socket-4wwvh5bx.livekit.cloud", "region": "India West", "protocol": 16} 113519.210 INFO livekit.agents closing agent session due to participant disconnect (disable via

RoomInputOptions.close_on_disconnect=False

) {"room": "voice_assistant_room_5981", "participant": "voice_assistant_user_6695", "reason": "CLIENT_INITIATED", "pid": 510204, "job_id": "AJ_bRujJQ6ihCxy", "room_id": "RM_UpQU5TXR6zr6"} 113519.590 INFO livekit.agents received job request {"job_id": "AJ_mpGGNgbXdLzQ", "dispatch_id": "AD_GztCs4NRUA58", "room": "voice_assistant_room_1526", "room_id": "RM_bqKpR7ukAufx", "agent_name": "", "resuming": false, "enable_recording": false} 113519.642 DEBUG livekit.agents received assignment {"agent_name": "", "room_id": "RM_bqKpR7ukAufx", "room": "voice_assistant_room_1526", "job_id": "AJ_mpGGNgbXdLzQ", "dispatch_id": "AD_GztCs4NRUA58", "enable_recording": false} 113519.648 INFO livekit.agents initializing process {"pid": 510405} 113520.104 INFO livekit.agents process initialized {"pid": 510405, "elapsed_time": 0.45} DEBUG asyncio Using selector: EpollSelector {"pid": 510405} 113520.162 DEBUG livekit.agents input stream attached {"participant": null, "source": "SOURCE_UNKNOWN", "accepted_sources": ["SOURCE_MICROPHONE"], "room": "voice_assistant_room_1526", "pid": 510405, "job_id": "AJ_mpGGNgbXdLzQ", "room_id": "RM_bqKpR7ukAufx"} VAAPI is supported. 113534.789 ERROR asyncio Task exception was never retrieved future: <Task finished name='Task-22' coro=<RoomIO._on_user_text_input.<locals>._read_text() done, defined at /home/karandeep/Development/agent-starter-python/.venv/lib/python3.10/site-packages/l… exception=RuntimeError('AgentSession is closing, cannot use generate_reply()')> Traceback (most recent call last): File "/home/karandeep/Development/agent-starter-python/.venv/lib/python3.10/site-packages/… line 408, in _read_text text_input_result = text_input_cb( File "/home/karandeep/Development/agent-starter-python/.venv/lib/python3.10/site-packages/… line 54, in _default_text_input_cb sess.generate_reply(user_input=ev.text) File "/home/karandeep/Development/agent-starter-python/.venv/lib/python3.10/site-packages/… line 941, in generate_reply raise RuntimeError("AgentSession is closing, cannot use generate_reply()") RuntimeError: AgentSession is closing, cannot use generate_reply() {"room": "voice_assistant_room_1526", "pid": 510405, "job_id": "AJ_mpGGNgbXdLzQ", "room_id": "RM_bqKpR7ukAufx"} Traceback (most recent call last): File "/home/karandeep/Development/agent-starter-python/.venv/lib/python3.10/site-packages/livekit/agents/voice/room_io/room_io.py", line 408, in _read_text text_input_result = text_input_cb( File "/home/karandeep/Development/agent-starter-python/.venv/lib/python3.10/site-packages/livekit/agents/voice/room_io/types.py", line 54, in _default_text_input_cb sess.generate_reply(user_input=ev.text) File "/home/karandeep/Development/agent-starter-python/.venv/lib/python3.10/site-packages/livekit/agents/voice/agent_session.py", line 941, in generate_reply raise RuntimeError("AgentSession is closing, cannot use generate_reply()") RuntimeError: AgentSession is closing, cannot use generate_reply()

Thanks — I’m responding about the AgentSession closing error and the MCP integration issues you sent. What’s happening • The RuntimeError ("AgentSession is closing, cannot use generate_reply()") occurs because the session is being closed (participant disconnected) while a text input handler calls

generate_reply()

. That prevents normal reply generation and can interrupt MCP tool load/use. • Your earlier curl showed the MCP endpoint issues (307 redirect to an SSE path). The redirect + SSE handling and the unpickling error you saw can all interfere with tool loading and session stability. What you can do now (quick checklist) • Prevent the session from auto-closing on participant disconnect: a. When starting the session, set

close_on_disconnect=False

in

RoomInputOptions

.

from livekit.agents.voice.room_io import RoomInputOptions

     await session.start(
         room=ctx.room,
         agent=Assistant(),
         room_options=RoomInputOptions(close_on_disconnect=False)
     )

• Verify MCP SSE redirect is reachable by the agent process: a. From the agent host run: curl -v "https://backend.composio.dev/v3/mcp/&lt;serverId&gt;/sse?transport=sse" 1. If that returns a 200 SSE stream, the agent should be able to connect. If it still redirects, confirm the final URL works and does not require special redirect handling. • Ensure the MCP optional dependency is present (you said it is installed) and versions are current: a. Share

pip show livekit-agents mcp

or

pip list | grep -E "livekit-agents|mcp"

if you want me to check compatibility. • Share MCP/server responses used by the agent: a. Paste the response for the

list_tools

request or run: curl -v "https://…/v3/mcp/<serverId>/list_tools" and paste the JSON if possible. • Address the unpickling AttributeError: a. That error suggests a mismatch in expected exception object shape during IPC. Please paste package versions (previous step) and any additional traceback lines around the unpickling error so I can point to a precise fix (often a version mismatch between

mcp

and

livekit-agents

). If you want, paste: • The agent log lines around the AgentSession closing and any MCP-related tracebacks, • Output of

pip list | grep -E "livekit-agents|mcp"

, • The curl result for the /sse path or the /list_tools response. I’ll use those to tell you the exact next change. Sources: Tool definition and use | LiveKit docs | Events and error handling | LiveKit docs

still the same error

Thanks — I hear you: the error persists. I’ll help narrow it down and get this fixed. Please provide the following so I can diagnose the root cause quickly: 1. The exact code where you create/start the AgentSession (including any

RoomInputOptions

and your

mcp_servers

config). Paste the AgentSession construction and the

await session.start(...)

call. 2. The agent package versions: ◦ Output of:

pip show livekit-agents mcp

(or

pip list | grep -E "livekit-agents|mcp"

). 1. MCP responses from the agent host: ◦ Result of: curl -v "https://backend.composio.dev/v3/mcp/&lt;serverId&gt;/sse?transport=sse" and curl -v "https://backend.composio.dev/v3/mcp/&lt;serverId&gt;/list_tools" • Paste the responses (headers + body/JSON). 1. Agent logs (full traceback) around the time of the error: ◦ Include ~20 lines before and after the RuntimeError and the earlier unpickling AttributeError. Quick checks you can run now (optional, helpful if you want to try before sharing logs): • Ensure

RoomInputOptions(close_on_disconnect=False)

is passed when starting the session. • Restart the agent after updating packages:

uv update livekit-agents[mcp] mcp

or

pip install -U livekit-agents[mcp] mcp

. • If you see the unpickling AttributeError again, search your environment for multiple versions of

mcp

or conflicting installs (virtualenv vs system). Send the code + package versions + curl outputs + log snippets and I’ll pinpoint the exact fix.

i'm using uv package manager and installed mcp through that still same error VAAPI is supported. 114600.873 ERROR asyncio Task exception was never retrieved future: <Task finished name='Task-22' coro=<RoomIO._on_user_text_input.<locals>._read_text() done, defined at /home/karandeep/Development/agent-starter-python/.venv/lib/python3.10/site-packages/l… exception=RuntimeError('AgentSession is closing, cannot use generate_reply()')> Traceback (most recent call last): File "/home/karandeep/Development/agent-starter-python/.venv/lib/python3.10/site-packages/… line 408, in _read_text text_input_result = text_input_cb( File "/home/karandeep/Development/agent-starter-python/.venv/lib/python3.10/site-packages/… line 54, in _default_text_input_cb sess.generate_reply(user_input=ev.text) File "/home/karandeep/Development/agent-starter-python/.venv/lib/python3.10/site-packages/… line 941, in generate_reply raise RuntimeError("AgentSession is closing, cannot use generate_reply()") RuntimeError: AgentSession is closing, cannot use generate_reply() {"room": "voice_assistant_room_6052", "pid": 519346, "job_id": "AJ_6mheXZvQSGLr", "room_id": "RM_CSExEBfgKMgh"} Traceback (most recent call last): File "/home/karandeep/Development/agent-starter-python/.venv/lib/python3.10/site-packages/livekit/agents/voice/room_io/room_io.py", line 408, in _read_text text_input_result = text_input_cb( File "/home/karandeep/Development/agent-starter-python/.venv/lib/python3.10/site-packages/livekit/agents/voice/room_io/types.py", line 54, in _default_text_input_cb sess.generate_reply(user_input=ev.text) File "/home/karandeep/Development/agent-starter-python/.venv/lib/python3.10/site-packages/livekit/agents/voice/agent_session.py", line 941, in generate_reply raise RuntimeError("AgentSession is closing, cannot use generate_reply()") RuntimeError: AgentSession is closing, cannot use generate_reply()

Thanks — I see the same RuntimeError repeating:

AgentSession is closing, cannot use generate_reply()

. That means the session is being closed (usually from participant disconnect) while your text input handler calls

generate_reply()

. To resolve this I need two quick confirmations and one small code change to test: 1) Confirm your AgentSession start code — paste the snippet where you construct AgentSession and call

await session.start(...)

. • I need to see if

RoomInputOptions(close_on_disconnect=False)

is being passed. 2) Confirm package versions (you said you used uv): • Run and paste:

pip show livekit-agents mcp

(or

pip list | grep -E "livekit-agents|mcp"

). 3) Temporary fix to test now — prevent auto-close: • If you aren’t already, start the session with:

from livekit.agents.voice.room_io import RoomInputOptions

     await session.start(
         room=ctx.room,
         agent=Assistant(),
         room_options=RoomInputOptions(close_on_disconnect=False)
     )

• Restart the agent and reproduce the error. If the RuntimeError disappears, the root cause is the session closing on disconnect. If it persists, paste agent logs (20 lines before/after the traceback). If you prefer, paste the AgentSession code now and I’ll point out exactly where to change it.

i added room_options=RoomInputOptions(close_on_disconnect=False), but still getting VAAPI is supported. 120347.079 ERROR asyncio Task exception was never retrieved future: <Task finished name='Task-22' coro=<RoomIO._on_user_text_input.<locals>._read_text() done, defined at /home/karandeep/Development/agent-starter-python/.venv/lib/python3.10/site-packages/livekit/agents/v… exception=RuntimeError('AgentSession is closing, cannot use generate_reply()')> Traceback (most recent call last): File "/home/karandeep/Development/agent-starter-python/.venv/lib/python3.10/site-packages/livekit/agents/… line 408, in _read_text text_input_result = text_input_cb( File "/home/karandeep/Development/agent-starter-python/.venv/lib/python3.10/site-packages/livekit/agents/… line 54, in _default_text_input_cb sess.generate_reply(user_input=ev.text) File "/home/karandeep/Development/agent-starter-python/.venv/lib/python3.10/site-packages/livekit/agents/… line 941, in generate_reply raise RuntimeError("AgentSession is closing, cannot use generate_reply()") RuntimeError: AgentSession is closing, cannot use generate_reply() {"room": "voice_assistant_room_2962", "pid": 534153, "job_id": "AJ_dXuvoFPxwYcp", "room_id": "RM_qaWDrnEjTxfr"} Traceback (most recent call last): File "/home/karandeep/Development/agent-starter-python/.venv/lib/python3.10/site-packages/livekit/agents/voice/room_io/room_io.py", line 408, in _read_text text_input_result = text_input_cb( File "/home/karandeep/Development/agent-starter-python/.venv/lib/python3.10/site-packages/livekit/agents/voice/room_io/types.py", line 54, in _default_text_input_cb sess.generate_reply(user_input=ev.text) File "/home/karandeep/Development/agent-starter-python/.venv/lib/python3.10/site-packages/livekit/agents/voice/agent_session.py", line 941, in generate_reply raise RuntimeError("AgentSession is closing, cannot use generate_reply()") RuntimeError: AgentSession is closing, cannot use generate_reply()

Thanks — I hear you: the error persists even after setting

close_on_disconnect=False

. I’ll help pinpoint why the session is still marked closing. Please paste these four items (they’re the minimum I need to diagnose): 1) The exact AgentSession construction + the

await session.start(...)

call you’re using (full snippet including imports and

room_options

). 2) The agent log lines that occur immediately before the RuntimeError (20–30 lines prior) — I need any messages that show why the session entered a closing state (participant disconnect, explicit close, or error). 3) Output of package versions: • Run and paste:

pip show livekit-agents mcp

(or

pip list | grep -E "livekit-agents|mcp"

). 4) The MCP-related curl outputs you ran earlier (final SSE URL and

list_tools

if available). If you still have the curl responses, paste them again. If you prefer, paste the full startup log and I’ll extract the relevant pre-error lines. Once I have those four items I’ll give a targeted fix.

see below: import logging from dotenv import load_dotenv from livekit import rtc from livekit.agents import ( Agent, AgentServer, AgentSession, JobContext, JobProcess, cli, inference, room_io, ) from livekit.plugins import noise_cancellation, silero from livekit.plugins.turn_detector.multilingual import MultilingualModel from livekit.agents import mcp from livekit.plugins import elevenlabs from livekit.plugins import openai from livekit.plugins import deepgram logger = logging.getLogger("agent") load_dotenv(".env.local") class Assistant(Agent): def __init__(self) -> None: super().__init__( instructions="""You are a helpful voice AI assistant. The user is interacting with you via voice, even if you perceive the conversation as text. You eagerly assist users with their questions by providing information from your extensive knowledge. Your responses are concise, to the point, and without any complex formatting or punctuation including emojis, asterisks, or other symbols. You are curious, friendly, and have a sense of humor.""", ) # To add tools, use the @function_tool decorator. # Here's an example that adds a simple weather tool. # You also have to add

from livekit.agents import function_tool, RunContext

to the top of this file # @function_tool # async def lookup_weather(self, context: RunContext, location: str): # """Use this tool to look up current weather information in the given location. # # If the location is not supported by the weather service, the tool will indicate this. You must tell the user the location's weather is unavailable. # # Args: # location: The location to look up weather information for (e.g. city name) # """ # # logger.info(f"Looking up weather for {location}") # # return "sunny with a temperature of 70 degrees." server = AgentServer() def prewarm(proc: JobProcess): proc.userdata["vad"] = silero.VAD.load() server.setup_fnc = prewarm @breezy-insurance-66772.rtc_session() async def my_agent(ctx: JobContext): # Logging setup # Add any other context you want in all log entries here ctx.log_context_fields = { "room": ctx.room.name, } # Set up a voice AI pipeline using OpenAI, Cartesia, AssemblyAI, and the LiveKit turn detector session = AgentSession( # Speech-to-text (STT) is your agent's ears, turning the user's speech into text that the LLM can understand # See all available models at https://docs.livekit.io/agents/models/stt/ stt=deepgram.STTv2( model="flux-general-en", eager_eot_threshold=0.4, ), # A Large Language Model (LLM) is your agent's brain, processing user input and generating a response # See all available models at https://docs.livekit.io/agents/models/llm/ llm=openai.LLM(model="gpt-4o-mini"), # Text-to-speech (TTS) is your agent's voice, turning the LLM's text into speech that the user can hear # See all available models as well as voice selections at https://docs.livekit.io/agents/models/tts/ # Default # tts=elevenlabs.TTS( # voice_id="ODq5zmih8GrVes37Dizd", # model="eleven_multilingual_v2" # ), # Eric # tts=elevenlabs.TTS( # model="eleven_turbo_v2_5", # voice_id="cjVigY5qzO86Huf0OWal", # ), # Alice tts=elevenlabs.TTS( model="eleven_turbo_v2_5", voice_id="Xb7hH8MSUJpSbSDYk0k2", ), # VAD and turn detection are used to determine when the user is speaking and when the agent should respond # See more at https://docs.livekit.io/agents/build/turns turn_detection=MultilingualModel(), vad=ctx.proc.userdata["vad"], # allow the LLM to generate a response while waiting for the end of turn # See more at https://docs.livekit.io/agents/build/audio/#preemptive-generation preemptive_generation=True, mcp_servers=[mcp.MCPServerHTTP(url="https://backend.composio.dev/v3/mcp/11a72dd6-fb32-4b58-a4d6-7936a6d9b48c")] ) # To use a realtime model instead of a voice pipeline, use the following session setup instead. # (Note: This is for the OpenAI Realtime API. For other providers, see https://docs.livekit.io/agents/models/realtime/)) # 1. Install livekit-agents[openai] # 2. Set OPENAI_API_KEY in .env.local # 3. Add

from livekit.plugins import openai

to the top of this file # 4. Use the following session setup instead of the version above # session = AgentSession( # llm=openai.realtime.RealtimeModel(voice="marin") # ) # # Add a virtual avatar to the session, if desired # # For other providers, see https://docs.livekit.io/agents/models/avatar/ # avatar = hedra.AvatarSession( # avatar_id="...", # See https://docs.livekit.io/agents/models/avatar/plugins/hedra # ) # # Start the avatar and wait for it to join # await avatar.start(session, room=ctx.room) # Start the session, which initializes the voice pipeline and warms up the models await session.start( agent=Assistant(), room=ctx.room, room_options=room_io.RoomOptions( close_on_disconnect=False, audio_input=room_io.AudioInputOptions( noise_cancellation=lambda params: noise_cancellation.BVCTelephony() if params.participant.kind == rtc.ParticipantKind.PARTICIPANT_KIND_SIP else noise_cancellation.BVC(), ), ), ) # Join the room and connect to the user await ctx.connect() if name == "__main__": cli.run_app(server)

Thanks — I see your full session code. You’re using

RoomOptions(close_on_disconnect=False)

(correct) and you still get repeated: RuntimeError: AgentSession is closing, cannot use generate_reply() This means something is still putting the session into a closing state while the text input task runs. Based on your logs, curl redirect, and prior unpickle error, the most likely causes and concrete next steps are: 1) Use the final SSE URL (follow the redirect) for the MCPServerHTTP URL • Your curl returned a 307 Location that ends with

/sse?...

. Use that final URL in

mcp.MCPServerHTTP(...)

so the agent connects directly to the SSE path instead of being redirected. Example:

mcp_servers=[mcp.MCPServerHTTP(url="<https://backend.composio.dev/v3/mcp/<serverId>/sse?transport=sse>")]

2) Confirm the MCP server returns a valid

list_tools

JSON on the expected endpoint • From the agent host run and paste results: ◦ curl -v "https://backend.composio.dev/v3/mcp/&lt;serverId&gt;/sse?transport=sse" ◦ curl -v "https://backend.composio.dev/v3/mcp/&lt;serverId&gt;/list_tools" • If

list_tools

is not valid JSON or is missing, tools won’t load and the session startup can fail. 3) Resolve the unpickling/exception mismatch (if still present) — check package versions and update • Run and paste results of: ◦ pip show livekit-agents mcp ◦ pip list | grep -E "livekit-agents|mcp" • If versions are mismatched, update with uv:

uv update livekit-agents[mcp] mcp

or with pip:

pip install -U livekit-agents[mcp] mcp

• Restart the agent after update. 4) Add a defensive check around generate_reply to avoid the RuntimeError while we diagnose • Replace the default callback or add a small guard before calling `sess.generate_reply(...)`:

if not sess.closing:
    await sess.generate_reply(user_input=ev.text)
else:
    <http://ctx.log.info|ctx.log.info>("Session closing - skipping generate_reply")

(The AgentSession API exposes the session state used internally; check your version for the exact attribute name — paste package versions and I’ll confirm.) 5) Collect logs that show why the session entered closing state • Paste ~30 lines before the RuntimeError (search for "closing", "disconnect", "error", "exception" in logs). Those lines show the triggering event (MCP connection failure, unhandled exception, participant disconnect, or explicit close). What I need from you next (small, concrete set) • Confirm you changed the MCP URL to the final

/sse

path and retested; paste the curl output for both

/sse

and

/list_tools

. • Paste

pip show livekit-agents mcp

(or

pip list

lines). • Paste ~30 log lines immediately before the RuntimeError. Once you paste those three items I’ll tell you the exact line to change (if any), or the package fix to apply, and whether the guard needs a different attribute name for your version.

here are logs: uv run python src/agent.py dev 121322.651 DEBUG asyncio Using selector: EpollSelector 121322.653 DEV livekit.agents Watching /home/karandeep/Development/agent-starter-python/src 121323.613 DEBUG asyncio Using selector: EpollSelector 121323.615 INFO livekit.agents starting worker {"version": "1.3.6", "rtc-version": "1.0.20"} 121323.616 INFO livekit.agents preloading plugins {"packages": ["livekit.plugins.silero", "livekit.plugins.turn_detector", "livekit.plugins.elevenlabs", "livekit.plugins.openai", "livekit.plugins.deepgram", "av"]} 121323.648 INFO livekit.agents starting inference executor 121324.257 INFO livekit.agents initializing process {"pid": 540901, "inference": true} 121326.841 INFO livekit.agents process initialized {"pid": 540901, "inference": true, "elapsed_time": 2.58} 121326.843 INFO livekit.agents HTTP server listening on :33555 121327.295 INFO livekit.agents registered worker {"agent_name": "", "id": "AW_s7ycBw6dAwoz", "url": "wss://twil-socket-4wwvh5bx.livekit.cloud", "region": "India West", "protocol": 16} 121341.990 INFO livekit.agents received job request {"job_id": "AJ_Nm8XYzkbVR9U", "dispatch_id": "AD_NcqrGxssgxtn", "room": "voice_assistant_room_3661", "room_id": "RM_SBVnLqhuDJ68", "agent_name": "", "resuming": false, "enable_recording": false} 121342.073 DEBUG livekit.agents received assignment {"agent_name": "", "room_id": "RM_SBVnLqhuDJ68", "room": "voice_assistant_room_3661", "job_id": "AJ_Nm8XYzkbVR9U", "dispatch_id": "AD_NcqrGxssgxtn", "enable_recording": false} 121342.078 INFO livekit.agents initializing process {"pid": 541116} 121342.572 INFO livekit.agents process initialized {"pid": 541116, "elapsed_time": 0.49} DEBUG asyncio Using selector: EpollSelector {"pid": 541116} 121342.657 DEBUG livekit.agents input stream attached {"participant": null, "source": "SOURCE_UNKNOWN", "accepted_sources": ["SOURCE_MICROPHONE"], "room": "voice_assistant_room_3661", "pid": 541116, "job_id": "AJ_Nm8XYzkbVR9U", "room_id": "RM_SBVnLqhuDJ68"} VAAPI is supported.

Thanks — I see the startup logs and that the session attaches input, but I still don’t see the lines that put the session into a closing state. To move this forward I need three targeted items (small, quick) so I can identify the exact trigger and give a precise fix: 1) Package versions (one command output) • Run and paste: pip show livekit-agents mcp 2) The MCP endpoint responses (two curl commands) • Run from the agent host and paste both full outputs (headers + body): ◦ curl -v "https://backend.composio.dev/v3/mcp/11a72dd6-fb32-4b58-a4d6-7936a6d9b48c/sse?transport=sse" ◦ curl -v "https://backend.composio.dev/v3/mcp/11a72dd6-fb32-4b58-a4d6-7936a6d9b48c/list_tools" 3) Agent log lines immediately before the RuntimeError • Paste ~30 lines preceding the RuntimeError traceback you already shared (the lines that show events leading up to "AgentSession is closing"). Those lines will show whether a disconnect, unhandled exception, MCP failure, or explicit close triggered the session shutdown. If you want a quick mitigation while we diagnose, add this guard around the generate call in your code where default callback triggers it (replace where

sess.generate_reply(...)

is invoked):

# quick defensive guard (adjust attribute name if your AgentSession uses a different flag)
if not getattr(sess, "closing", False) and not getattr(sess, "closed", False):
    await sess.generate_reply(user_input=ev.text)
else:
    <http://ctx.log.info|ctx.log.info>("Session closing/closed — skipping generate_reply")

Send the three items above and I’ll pinpoint the root cause and exact code/package change needed.

there's no existng generate_reply in my codebase

Thanks — I understand you don’t call

generate_reply()

in your code. That’s expected: the default text input callback inside the LiveKit Agents library calls

generate_reply()

for you, which is where the RuntimeError originates. What to do next (minimal, actionable): 1) Replace the default text input callback with a small custom callback that guards against session closing. Add this near your imports and pass it into

room_io.RoomOptions

as `text_input_cb`:

from livekit.agents.voice import room_io

async def safe_text_input_cb(sess, ev, ctx):
    # sess is the AgentSession used by the callback
    if getattr(sess, "closing", False) or getattr(sess, "closed", False):
        <http://ctx.log.info|ctx.log.info>("Session closing/closed — skipping generate_reply")
        return None
    # call the existing default behavior
    return await room_io.types._default_text_input_cb(sess, ev, ctx)

Then in your

session.start(...)

call use:

room_options=room_io.RoomOptions(
    close_on_disconnect=False,
    text_input_cb=safe_text_input_cb,
    ...
)

2) Send me your package versions so I confirm the correct session-state attribute name: • Run and paste: pip show livekit-agents mcp 3) If you want me to produce a drop-in replacement for your library version, paste the two things above and I’ll return the exact callback code aligned to your package versions. Do you want me to generate the ready-to-paste callback adjusted to your livekit-agents version?