Executor version
v2
How do you run Executor?
Executor Cloud
Operating system
Linux
Integration involved
OpenAPI
What happened
My codex instance keep getting disconnected from executor MCP.
I think this is third-time since yesterday.
Environment
- Date: 2026-10-01. All times below are UTC
- Endpoint: https://v2.executor.sh/mcp
- Client: T3 Code using Codex App Server, Codex CLI version 0.159.2.
- Authentication: browser OAuth; organization-bound hosted connection.
- Dawarich API key remained valid; the same saved account worked after Executor reauthentication, without changing the Dawarich key.
Sol 6.1 tought it was somehow related to call limits but couldn't reproduce.
So adding timeline generated by codex in case needs investigation on server end.
Also redirect to localhost auth needs an alternative. Trying to auth from ssh session becomes painful.
What you expected
Keep me logged-in to executor.
Steps to reproduce
Diagnostics / logs
Observed timeline
- 09:42:37.264: Executor MCP startup status ready.
- 09:54:06.833: Executor MCP status ready, no startup error.
- 09:55:24.127: A 100-place deletion execution completed successfully (10,679 ms).
- 09:55:45.282: The next, larger execution returned 100 successful deletions; remaining attempts hit "Execution exceeded its tool-call limit of 100." This is an expected documented execution limit and a separate batching mistake, not proof of a crash. Exactly 200 deletions had succeeded across the two executions.
- 09:55:59.178: Next execute call failed with "Authentication required" (499 ms).
- 09:56:04.109: Minimal tools.search retry failed identically (594 ms).
- 09:57:25.852: Client still reported Executor MCP status ready, with no startup error.
- 09:57:34.156: Another minimal tools.search failed identically (519 ms).
- 09:58:44.017 to 09:58:47.152: Executor MCP changed from starting to ready around manual reauthentication.
- 09:58:57.536: Brief client tool availability race: functions.exec reported "tools.mcp__executor__execute is not a function" despite the name remaining in ALL_TOOLS. Rediscovery and retry resolved this without another sign-in. This is a separate client-side symptom.
- 09:59:15.199: Executor tools.search succeeded again (1,320 ms).
- Subsequently: All 1,930 remaining deletions completed using separate executions of at most 100 calls, followed by a successful read-only verification. No further reauthentication was needed during that cleanup.
Before you submit
Executor version
v2
How do you run Executor?
Executor Cloud
Operating system
Linux
Integration involved
OpenAPI
What happened
My codex instance keep getting disconnected from executor MCP.
I think this is third-time since yesterday.
Sol 6.1 tought it was somehow related to call limits but couldn't reproduce.
So adding timeline generated by codex in case needs investigation on server end.
Also redirect to localhost auth needs an alternative. Trying to auth from ssh session becomes painful.
What you expected
Keep me logged-in to executor.
Steps to reproduce
Diagnostics / logs
Before you submit