fal MCP "why did my request fail": the Sume job equivalent

fal's Platform MCP server is for asking an assistant "why did my last request to my-app fail?" about your fal account. Sume has the same kind of loop for media jobs: from hosted MCP, an agent reads jobs_status, jobs_events, and jobs_result for the job id, then retries according to the error.

The fal facts are from its changelog (Aug 17 entry, read 2026-10-01). This compares the workflow, not the features one for one. Sume details are from Jobs and results.

What does the fal Platform MCP do?

The changelog describes a second MCP server at https://api.fal.ai/v1/mcp/platform that lets an assistant operate your fal account. It lists 15 tools: eleven serverless debugging tools covering requests, logs, error analytics, deploy and revision history, runner and queue state, spend, and storage, plus a four-tool discovery gateway. It is read-only, stateless, and free. Setup is in fal MCP server setup and billing.

How do I ask the same question about a Sume job?

Give the agent the job id. The hosted MCP job read tools are jobs_list, jobs_get, jobs_status, jobs_result, jobs_events, and jobs_wait. Over HTTP the same timeline is GET /v1/jobs/{id}/events, a pull snapshot rather than a stream.

What should the agent do once it knows why?

Follow the error envelope rules. Back off on 429 and use retry-after when present; queue_full is not ordinary rate limiting and clears when an existing job finishes or is canceled. Do not retry unsafe submits without an Idempotency-Key. The public events do not expose raw provider task ids or URLs, so the answer comes from the Sume error, not provider internals.

Does a failure mean I was billed?

The docs say a client timeout does not cancel a job and it still bills. For a terminal failure, read the job's error and your usage records before assuming either way; see errors and credits.