check_image_jobs
Batch-poll multiple async image jobs in one call.
readSellerSheet
Use this when you have submitted several edit_image jobs in parallel (e.g. a family-recolor fan-out) — avoids N separate round trips per polling cycle. Rows stay small (~1 KB per job) and carry no image. Each row already carries cdn_url; call check_image_job only for a 'done' job whose cdn_url is still null.
Call it
In an AI client (MCP)
{
"name": "check_image_jobs",
"arguments": {
"job_ids": []
}
}From code (Tool API)
curl -X POST https://api.sellersheetai.com/api/v1/tools/check_image_jobs \
-H "Authorization: Bearer $SELLERSHEET_API_KEY" \
-H "Content-Type: application/json" \
-d '{"job_ids":[]}'Early access — the Tool API is open on request, contact us.
Arguments
| Name | Type | Description |
|---|---|---|
job_idsrequired | object[] | up to 20 job_ids from edit_image; ids past the 20th are dropped without notice. |
Result
Every tool answers {notification, data, human_action}; data carries:
data.jobs— one row per id, in input order — id, status ('pending' | 'processing' | 'done' | 'error' | 'not_found' = unknown or past its ~15-minute retention), progress, elapsed_s, queue_wait_s, cdn_url, error.data.done / data.pending / data.error— counts — poll until data.pending is 0.
Workflow
- 1. Submit N jobs via edit_image → collect job_ids 2. Poll check_image_jobs(job_ids) every 20–30s 3. When data.pending is 0, read cdn_url from each 'done' row.
Gotchas
- Successes = data.done (or cdn_url); failures = data.error only, not slow jobs. Re-submit only 'error' jobs, never a 'pending'/'processing' one (that duplicates and rebills).