I’m having an issue with integrating an MCP server during testing. To try and simplify this I am at the point of the simplest schema and request for it and I’ve removed all our code and am just testing through the API playground and get the same results. The server is HTTPS (but not on 443). The endpoint is SSE. The call to list tools works fine, it then makes a call to the simple tool with correct query params. I can see on our server the response is returned and processed gracefully with no disconnections. The API then returns a tool failure: 424: unhandled errors in a TaskGroup (1 sub-exception) This server works with MCP Inspector and also works when attached to Claude. Attempted on a few models. The only weird thing in all this I can think of is not using 443 - but the MCP is called, response is returned and only then does error happen. For sake of clarity here is the request/response schema - I’ve gone from a set of super simple tools to “about as simple as it gets” and we’re still getting the issue. { "type": "object", "properties": { "result": { "type": "string", "nullable": true } }, "$schema": "http://json-schema.org/draft-04/schema#", "additionalProperties": false } { "result": "You searched for: test" } Auth is bearer token (and working/provided fine). SSL cert is fine, connection is public, no connection resets, SSE is fine for everything else, it’s got minimal proxying (just border SSL offload) and no buffering or anti-SSE HTTP patterns inbetween. [image] Attached view of playground. Full stack of API calls: { "type": "response.created", "sequence_number": 0, "response": { "id": "resp_68a756202cb08196aa37b4d8eb91b5250153f9a2f86d22ba", "object": "response", "created_at": 1755797024, "status": "in_progress", "background": false, "error": null, "incomplete_details": null, "instructions": null, "max_output_tokens": null, "max_tool_calls": null, "model": "gpt-4o-2024-08-06", "output": [], "parallel_tool_calls": true, "previous_response_id": null, "prompt_cache_key": null, "reasoning": { "effort": null, "summary": null }, "safety_identifier": null, "service_tier": "auto", "store": true, "temperature": 1, "text": { "format": { "type": "text" }, "verbosity": "medium" }, "tool_choice": "auto", "tools": [ { "type": "mcp", "allowed_tools": null, "headers": { "Authorization": " " }, "require_approval": "never", "server_description": null, "server_label": "Orchestrator", "server_url": "https://UrlRedacted/ " } ], "top_logprobs": 0, "top_p": 1, "truncation": "disabled", "usage": null, "user": null, "metadata": {} } }, { "type": "response.in_progress", "sequence_number": 1, "response": { "id": "resp_68a756202cb08196aa37b4d8eb91b5250153f9a2f86d22ba", "object": "response", "created_at": 1755797024, "status": "in_progress", "background": false, "error": null, "incomplete_details": null, "instructions": null, "max_output_tokens": null, "max_tool_calls": null, "model": "gpt-4o-2024-08-06", "output": [], "parallel_tool_calls": true, "previous_response_id": null, "prompt_cache_key": null, "reasoning": { "effort": null, "summary": null }, "safety_identifier": null, "service_tier": "auto", "store": true, "temperature": 1, "text": { "format": { "type": "text" }, "verbosity": "medium" }, "tool_choice": "auto", "tools": [ { "type": "mcp", "allowed_tools": null, "headers": { "Authorization": " " }, "require_approval": "never", "server_description": null, "server_label": "Orchestrator", "server_url": "https://UrlRedacted/ " } ], "top_logprobs": 0, "top_p": 1, "truncation": "disabled", "usage": null, "user": null, "metadata": {} } }, { "type": "response.output_item.added", "sequence_number": 2, "output_index": 0, "item": { "id": "mcpl_68a75620853c8196a49404114463ba7c0153f9a2f86d22ba", "type": "mcp_list_tools", "server_label": "Orchestrator", "tools": [] } }, { "type": "response.mcp_list_tools.in_progress", "sequence_number": 3, "output_index": 0, "item_id": "mcpl_68a75620853c8196a49404114463ba7c0153f9a2f86d22ba" }, { "type": "response.mcp_list_tools.completed", "sequence_number": 4, "output_index": 0, "item_id": "mcpl_68a75620853c8196a49404114463ba7c0153f9a2f86d22ba" }, { "type": "response.output_item.done", "sequence_number": 5, "output_index": 0, "item": { "id": "mcpl_68a75620853c8196a49404114463ba7c0153f9a2f86d22ba", "type": "mcp_list_tools", "server_label": "Orchestrator", "tools": [ { "annotations": { "read_only": false }, "description": "Search for a file, just enter the query and get some results!", "input_schema": { "type": "object", "properties": { "query": { "type": "string", "description": "The query to use to search documents for.", "minLength": 1 } }, "required": [ "query" ], "additionalProperties": false }, "name": "FileSearch" } ] } }, { "type": "response.output_item.added", "sequence_number": 6, "output_index": 1, "item": { "id": "mcp_68a75636e7408196a692b078b82ecb360153f9a2f86d22ba", "type": "mcp_call", "approval_request_id": null, "arguments": "", "error": null, "name": "FileSearch", "output": null, "server_label": "Orchestrator" } }, { "type": "response.mcp_call.in_progress", "sequence_number": 7, "output_index": 1, "item_id": "mcp_68a75636e7408196a692b078b82ecb360153f9a2f86d22ba" }, { "type": "response.mcp_call_arguments.delta", "sequence_number": 8, "output_index": 1, "item_id": "mcp_68a75636e7408196a692b078b82ecb360153f9a2f86d22ba", "delta": "{\"query\":\"test\"}", "obfuscation": "" }, { "type": "response.mcp_call_arguments.done", "sequence_number": 9, "output_index": 1, "item_id": "mcp_68a75636e7408196a692b078b82ecb360153f9a2f86d22ba", "arguments": "{\"query\":\"test\"}" }, { "type": "response.mcp_call.failed", "sequence_number": 10, "output_index": 1, "item_id": "mcp_68a75636e7408196a692b078b82ecb360153f9a2f86d22ba" }, { "type": "response.output_item.done", "sequence_number": 11, "output_index": 1, "item": { "id": "mcp_68a75636e7408196a692b078b82ecb360153f9a2f86d22ba", "type": "mcp_call", "approval_request_id": null, "arguments": "{\"query\":\"test\"}", "error": { "type": "http_error", "code": 424, "message": "424: unhandled errors in a TaskGroup (1 sub-exception)" }, "name": "FileSearch", "output": null, "server_label": "Orchestrator" } }

Hi, I’m having the same problem. Have you found a solution? Alexandre

No, I had to end up translating MCP into tool calls to get this to work at all.

I finally found the issue. I defined an “output schema” but the structuredContent was missing in the response. I saw it with the tool “MCP inspector”

I’ve fixed this by returning structured data from the tool instead of text.

I was able to solve this problem, and posted my solution in another thread reporting this issue. TLDR: The issue was MCP resource discovery. ChatGPT calls resources/list while initializing or using a connector. My server registered tools but did not register any resources. Adding one valid, readable resource fixed every tool call immediately: Link to my longer writeup: [image] MCP Integration unhandled errors in a TaskGroup (1 sub-exception) Codex I was able to solve this problem, so here’s a solution for posterity. I hit the same generic error when calling tools through ChatGPT: ExceptionGroup: unhandled errors in a TaskGroup (1 sub-exception) The MCP server itself was healthy, the Secure MCP Tunnel was connected, and ChatGPT could discover the connector and its tools. However, every tool call failed. ChatGPT sometimes presented this as an expired connection, which made it initially look like an authentication or tunnel problem. The …

Hey @guytp, sorry for the late reply. Thanks for the detailed trace and the testing you’ve already done. Are you still seeing this error? Thanks also to @alexandre.bertrand, @guicalabria, and @somedude for sharing what resolved the error in your own setups. Those follow-ups are helpful for others finding this thread. If it’s still happening and your tool declares an outputSchema, check that its returned structuredContent matches that schema. OpenAI’s MCP testing guide includes this check. If the error continues, please add a recent request ID, failure timestamp with time zone, and a small request/response example to your existing private Support request. Remove credentials and private information before sharing the example. If you haven’t opened a request, use the chat bubble at help.openai.com. ~AD

the issue turned out to be the failure of an incompetently deployed oauth upgrade where the “rollback” made matters worse not better [image] Service unavailabilty python failure again Bugs yyep. looks like a massive screw-up by OpenAI - a repeat of a few months back. am currently getting “service unavailable” on the app. what i suspect is going on is that just as last time a container upgrade was rolled out, it effed up, and the only “fix” is to take the entire service offline and do a manual rollback. guys: i have told you, you need service API revision control. it is dead-easy to achieve via an intelligent HTTP proxy: the proxy looks at the version number of the API from the …

Thanks for linking the detailed report, @locl. I understand your concern about clearer updates on what failed and how recovery was handled. Are you still running into these tool failures? If so, please add a recent example to your existing private Support request, including the steps to reproduce it, the exact error, and the failure time with time zone. If you already have a feedback ID for that failure, include it there too. If you haven’t opened a Support request, use the chat bubble at help.openai.com. Please share only the Support case number here and keep account details, conversation links, screenshots and logs in the private request. ~AD

API Returns 424 Error for MCP Calls

API Bugs

somedude August 24, 2026, 2:43pm 7

I was able to solve this problem, and posted my solution in another thread reporting this issue.

TLDR:

The issue was MCP resource discovery. ChatGPT calls resources/list while initializing or using a connector. My server registered tools but did not register any resources. Adding one valid, readable resource fixed every tool call immediately:

Link to my longer writeup:

Topic		Replies	Views
Error when attempting to connect my Remote MCP in new Agent Builder API mcp , agent-builder	55	4359	December 1, 2025
Anyone else hitting 424 / TaskGroup errors when creating a custom MCP connector? Bugs mcp	8	1815	October 18, 2026
ChatGPT Custom MCP OAuth fails to get tools when auth enabled Codex chatgpt , api , mcp , connector	6	3727	September 21, 2025
MCP Connector "Resource not found" - tools/call never reaches server ChatGPT Apps SDK bug	32	2557	July 19, 2026
Custom MCP connector: web client drops the stream mid-turn ("Connection interrupted"), works fine on mobile — same server, same connector ChatGPT Apps SDK apps-sdk	8	1704	August 18, 2026

API Returns 424 Error for MCP Calls

Related topics