Moxn owl-glass MCP down again: Bad Request on reconnect, third occurrence today #1084

New issue

Closed

opened 2026-08-19 18:10:14 +00:00 by coilyco-ops-gaming · 2 comments

coilyco-ops-gaming commented

2026-08-19 18:10:14 +00:00

Member

The moxn tools are still listed on the Dowel lane but every call fails at the transport layer, the same signature as the two earlier outages today.

Observed this turn (2026-08-19 ~18:10Z): moxn.find and moxn.search both failed with upstream MCP session is closed (reconnect also failed: reconnect upstream MCP "https://owl-glass.moxn.dev/api/mcp/http": sending "notifications/initialized": Bad Request). The failure happens before any tool call, so nothing was read or written.

This is the documented hand-minted credential expiry mode: tools stay listed, nothing renews the credential, all calls fail until a human mints a new one. It matches #1026 (closed 04:46Z) and #1038 (closed 04:45Z) exactly, so the earlier re-mint did not stick or has already expired again.

Ask: re-mint / renew the owl-glass moxn credential and confirm a read call returns clean. The recurring pattern suggests the renewal itself needs a check or the credential lifetime is too short.

The moxn tools are still listed on the Dowel lane but every call fails at the transport layer, the same signature as the two earlier outages today. Observed this turn (2026-08-19 ~18:10Z): moxn.find and moxn.search both failed with `upstream MCP session is closed (reconnect also failed: reconnect upstream MCP "https://owl-glass.moxn.dev/api/mcp/http": sending "notifications/initialized": Bad Request)`. The failure happens before any tool call, so nothing was read or written. This is the documented hand-minted credential expiry mode: tools stay listed, nothing renews the credential, all calls fail until a human mints a new one. It matches #1026 (closed 04:46Z) and #1038 (closed 04:45Z) exactly, so the earlier re-mint did not stick or has already expired again. Ask: re-mint / renew the owl-glass moxn credential and confirm a read call returns clean. The recurring pattern suggests the renewal itself needs a check or the credential lifetime is too short.

coilyco-ops-gaming added the

🔒⚠️📦⚠️🔒 SANDBOXED 🔒⚠️📦⚠️🔒

label

2026-08-19 18:10:14 +00:00

coilyco-ops-gaming commented

2026-08-19 18:12:12 +00:00

Author

Member

Diagnosis correction, per review: the 400 Bad Request signature is NOT an auth failure. Auth failures are 401/403 - and #1038's real signature was sending "initialize": Unauthorized (401), which does not match today. Only #1026 shared the 400 shape (tools/list / notifications/initialized), and its "credential expiry" diagnosis was itself assumed, not proven. A 400 on notifications/initialized - a protocol handshake notification sent after initialize - means the server rejected the request shape or session state, not the credential. Re-verified 18:11Z: same sending "notifications/initialized": Bad Request, still before any tool call.

Revised ask: inspect the MCP transport/session handling on the owl-glass endpoint (protocol version, headers, session lifecycle) in addition to the credential. The ~04:45 re-mint correlated with recovery of both issues, but a fresh session would clear a stale-session 400 just as well, so the correlation does not identify the cause.

Diagnosis correction, per review: the 400 Bad Request signature is NOT an auth failure. Auth failures are 401/403 - and #1038's real signature was `sending "initialize": Unauthorized` (401), which does not match today. Only #1026 shared the 400 shape (`tools/list` / `notifications/initialized`), and its "credential expiry" diagnosis was itself assumed, not proven. A 400 on `notifications/initialized` - a protocol handshake notification sent after initialize - means the server rejected the request shape or session state, not the credential. Re-verified 18:11Z: same `sending "notifications/initialized": Bad Request`, still before any tool call. Revised ask: inspect the MCP transport/session handling on the owl-glass endpoint (protocol version, headers, session lifecycle) in addition to the credential. The ~04:45 re-mint correlated with recovery of both issues, but a fresh session would clear a stale-session 400 just as well, so the correlation does not identify the cause.

coilyco-ops-gaming commented

2026-08-19 18:30:34 +00:00

Author

Member

Confirmed recovered. At 2026-08-19 18:30Z a moxn.find read returned cleanly (5 documents, no transport error), where earlier this turn the same call shape failed on notifications/initialized: Bad Request before any tool call ran. The failure signature has cleared since filing, consistent with the analysis that the 400 was a server-side handshake/session-state problem rather than the credential (an expired bearer token would 401/403). No re-mint was performed between filing and this recovery, which further supports a restart/redeploy clearing the real cause. Closing as resolved.

Confirmed recovered. At 2026-08-19 18:30Z a `moxn.find` read returned cleanly (5 documents, no transport error), where earlier this turn the same call shape failed on `notifications/initialized: Bad Request` before any tool call ran. The failure signature has cleared since filing, consistent with the analysis that the 400 was a server-side handshake/session-state problem rather than the credential (an expired bearer token would 401/403). No re-mint was performed between filing and this recovery, which further supports a restart/redeploy clearing the real cause. Closing as resolved.

coilyco-ops-gaming closed this issue

2026-08-19 18:30:34 +00:00