VoltAgent MCP 30000 ms default timeout vs Sume jobs_wait

VoltAgent's docs give each MCP server an optional timeout in milliseconds with a default of 30000, and an example that sets 15000. Sume's jobs_wait holds one call for 50 seconds by default and 55 at most, so with the default timeout a wait on a render that is still running can outlast VoltAgent's limit. Set timeout above 55000, for example 60000, on the Sume entry.

Read on 2026-10-06: VoltAgent's MCP docs and Sume's Jobs and results.

What is the right wait pattern?

Do not request a longer wait; the hosted server clamps timeout_seconds to 55 and reports it in wait_slice_clamped. Wait in slices instead: call jobs_wait, and when it returns wait_slice_expired, call it again with the same ids. Sume's docs say a ten-minute render is waited out by repeating the call.

• Store the job id from the create response before the first wait.
• Never resubmit the paid create because a wait ended; reuse the same idempotency_key if the create itself must be retried.
• Treat a 524, 522, 523 or 525 on jobs_wait as transport, not as a job result.
• Call disconnect() on the configuration only when you are done; Sume jobs keep running after a disconnect.

Who decides when to stop?

The model loops on jobs_wait, so cap the loop in your agent: a step limit or a total time budget. A job that runs past your budget is still running and billing; cancel it with jobs_cancel while it is before generation, since Sume's docs say cancellation succeeds only before generation work starts.