Support MCP( v2 ) Tasks extension in ChatGPT Web

ChatGPT Feature requests
1

Title: Support MCP Tasks extension in ChatGPT Web

Now that MCP Tasks is an official extension (io.modelcontextprotocol/tasks), it would be great to see native support in the ChatGPT Web MCP client, especially for long-running tool calls through Secure MCP Tunnel.

2

ChatGPT Developer Mode already provides strong MCP support, but long-running MCP operations still require custom polling/workarounds.

The current MCP specification now provides a standardized solution through the Tasks Extension (SEP-2663) in MCP 2026-07-28.

Could ChatGPT support:

  • io.modelcontextprotocol/tasks capability negotiation
  • task-returning tools/call
  • tasks/get
  • tasks/update
  • tasks/cancel
  • task status notifications
  • durable task state across long-running operations and reconnects

This would let MCP servers start long-running work without keeping a tool call open or requiring each app/server to invent its own polling protocol.

Official Tasks overview:
modelcontextprotocol.io/extensions/tasks/overview

Related discussions already exist here:

  • Handling Long-Running Tasks in ChatGPT Apps — explicitly discusses waiting for MCP Tasks support:
    community.openai.com/t/handling-long-running-tasks-in-chatgpt-apps/1369488
  • Progress Notifications Not Working in ChatGPT — @OpenAI_Support previously discussed task-based patterns for long-running MCP operations:
    community.openai.com/t/progress-notifications-not-working-in-chatgpt-mcp-ts-sdk-1-20-0/1367559/2

SEP-2663 and MCP 2026-07-28 now provide a concrete standard for this problem. Is support for the Tasks Extension planned for ChatGPT Developer Mode / Apps SDK MCP hosts?

@OpenAI_Support

3

Yes this would be pretty awesome to have. Currently we are all having to get around this by making get_result, with a wait, or pool tools, just to get a long running response

4

Thanks for explaining the MCP Tasks workflow, including task status, cancellation and reconnect support. We'll pass this specific request to the team. We don't have a timeline to share, but appreciate the detail!