-
Notifications
You must be signed in to change notification settings - Fork 0
Pitfall ChatGPT Plus MCP Availability
On the ChatGPT Plus account tested on 2026-08-20, a custom connector completes
OAuth and authenticated MCP initialize, but ChatGPT sends neither
notifications/initialized nor tools/list. The UI reports that no actions
are available.
ChatGPT remains an MCP-compatible client. This page records a plan-specific, dated availability result; it does not claim that Pro, Business, Enterprise, or Edu accounts fail.
- The same behavior was reproduced with
read,write, andadmin, including a strictreadclient using?profile=reader. - Static OAuth and fresh DCR clients, rotated secrets, and real deployments of v1.8.8, v1.9.0, v1.9.1, and v1.9.2 all reached the same boundary.
- The unchanged v1.9.1 binary had completed a real ChatGPT tool call on 2026-08-17, then failed without a server release from 2026-08-20 onward.
- MCPJam completed discovery, DCR, PKCE, token exchange,
initialize, andtools/listagainst the same deployment and loaded 32 read tools. - Claude.ai also remained operational against the server.
This isolates the incident from server version, credentials, granted scope, Cloudflare, and the MCP/OAuth implementation. Server logs cannot distinguish a ChatGPT plan restriction from another client-side eligibility check.
Two OpenAI pages disagree as of 2026-08-20:
- The developer-mode guide lists Pro, Plus, Business, Enterprise, and Education accounts as eligible.
- The Help Center availability article limits full MCP to Business/Enterprise/Edu, describes Pro as read/fetch-only, and does not list Plus.
The second page is the current plan-availability warning. The first page is kept here because the contradiction is itself relevant diagnostic evidence.
- Verify that
/authorize,/token, and MCPinitializesucceed. - Check whether
notifications/initializedandtools/listfollow. - Run the same endpoint through a neutral MCP client such as MCPJam.
- Record the ChatGPT plan and test date; do not generalize one plan's result.
- If the control client works, report the evidence to OpenAI Support.
Do not loosen redirect URI validation, grant broader scopes, rotate working credentials repeatedly, or roll back the server solely for this symptom.
Mark the affected plan usable again only after a real ChatGPT reconnect reaches
tools/list and completes a representative tool call. Keep ChatGPT in the
compatibility matrix throughout; update only the dated plan-specific status.