Cloud Agents stuck on "Planning next moves" — env running, zero model output (2026-09-03)

Support Bug Reports
1

Where does the bug appear (feature/product)?

Cloud Agent (GitHub, Slack, Web, Linear)

Describe the Bug

Since ~13:00–15:00 Asia/Shanghai on 2026-09-03, new Cloud Agents on my account create successfully and reach running, but model inference never starts. UI stays on “Planning next moves”. No first assistant message, no tool calls.

Not a frontend-only hang: API status is running, then later error or finished, but the dump is only the original user prompt (0 assistant / 0 reasoning / 0 tool calls).

Not one session, not one model, not one repo. Stopping and relaunching hangs then errors. Same test on a different connected repo also hangs then errors. Same account previously completed Cloud Agent runs and opened PRs — recent regression.

The official Cursor status page showed All Systems Operational, Cloud Agents operational, no open incident. t.

Failure point: after environment ready, before the first model token.

Steps to Reproduce

  1. Start a new Cloud Agent from cursor.com/agents or the IDE Cloud dropdown on a connected repo.
  2. Use any model (reproduced across models; not Auto-only).
  3. Wait after the environment becomes ready.

Expected Behavior

After the environment is ready, the agent produces a first assistant message and starts tool calls.

Operating System

Other

Version Information

Cloud Agent backend issue, not IDE/CLI/iOS-version-specific.
Reproduced from the web Cloud Agents UI on newly created cloud sessions.

For AI issues: which model did you use?

Multiple models. Same hang regardless of model.

For AI issues: add Request ID with privacy disabled

bc-f36786f0-7fc3-47df-9b46-5eaae6731639
bc-d6ff95c4-8fa9-4dd0-ab09-60fcf239a4e1
bc-9edb9b21-19bc-4504-91da-fa55761d2172
bc-0997c2ce-24cc-45ce-8491-e4cc3cc9a446

Additional Information

bc-f36786f0 later marked finished, dump is a single user line (~95 bytes), no assistant output.
bc-d6ff95c4, bc-9edb9b21, bc-0997c2ce all ended error with empty transcripts.
IDs are lookup keys only; no prompt/repo/transcript pasted.

Does this stop you from using Cursor

Yes - Cursor is unusable

5

Hey @qi_zhao, thanks for the detailed report and the agent IDs; they made this easy to trace!

Your agents are starting fine, but the very first model call never leaves our servers, and every affected run happens after new MCP servers were added to your Cloud Agent configuration this morning. Runs before that change on the same account streamed normally. So this looks like one of the newly added servers is stalling the agent before its first response.

Could you try this so we can pin down which one:

  1. Open cursor.com/agents and go to Settings, then MCP.
  2. Disable (or remove) all of your MCP servers.
  3. Start a brand-new Cloud Agent on any connected repo with any model.
  4. If it now produces a first message, re-enable your servers one at a time, starting a brand-new agent after each one, until the hang comes back. The last server you enabled before it hung is the culprit.
  5. If step 3 still hangs with everything disabled, let us know; that would point to something other than MCP!
7

Update: reproduced and isolated.

I had not added MCP from the Cloud Agents UI. This morning two custom HTTP MCP servers were added on the same Cursor account from Grok Bot (AddMcpServer), and they show up in the Agents + → MCP Servers list:

  • aliyun-mcp-core (connected)
  • aliyun-openapi (Error)

Enabling either one is enough to hang new Cloud Agents on “Planning next moves” (env running, no first model token). Disabling both restores a fresh Cloud Agent to first-token output.

Gmail / Google Calendar / Google Drive can stay on; they do not cause the hang.

So the stall is account-level Aliyun MCP discovery/auth on first tool load, not a Cloud Agents platform outage. Happy to leave those two off.