Agent SDK rewind_conversation and a Sume job: find, then cancel
After rewind_conversation or a priority-now message, the MCP call is handled client-side. A submitted Sume job has its own id: find it with jobs_list.
Agent SDK 0.3.285 stops rewind_conversation from leaving a backgrounded MCP tool call running once its starting message is removed. That governs the client's call. A Sume job already submitted has its own id, so look it up with jobs_list and decide whether to call jobs_cancel.
SDK facts are from its release notes; job tools from MCP tools and gates, read 2026-10-01. The docs do not say a client-side cancel propagates to Sume, so check rather than assume.
What did the two releases change?
0.3.285 (2026-09-29): fixed rewind_conversation leaving a backgrounded MCP tool call running after the message that started it was removed. 0.3.286 (2026-09-30): a person's "priority now" message now moves running shell commands, agents and MCP calls to the background and joins the running turn instead of stopping it.
How do I find a Sume job after a rewind?
Use the read tools; none of them need an idempotency_key.
| Tool | Kind | Use after a rewind |
|---|---|---|
jobs_list | Read | See what is still in flight |
jobs_get, jobs_status | Read | Check one job |
jobs_result | Read | Collect artifacts if it finished |
jobs_events | Read | Review what happened |
jobs_cancel | Write | Needs idempotency_key and write scope |
When should I call jobs_cancel?
Only for a job you no longer want. It is a write tool, so it needs mcp:write or an API key; a read-only OAuth session gets insufficient_scope. Do not resubmit the original create just because the turn was rewound, since that would start a second paid job. Related: interrupts and running jobs.
What should I keep from the rewound turn?
The job id, and later the result. The docs advise reporting Sume public ids and media.sume.com URLs rather than pasting signed URLs, tokens or keys into chat logs.
Sources
Related posts
More in Developers
- Claude Code: disable an MCP server mid-session, Sume jobs persist
Switching off an MCP server in Claude Code SDK sessions removes its tools for new calls. Sume jobs already submitted keep running until you read or cancel them.
- Claude API compaction block: keep Sume job ids past the summary
The compact-2026-09-04 beta swaps old messages for a signed compaction block. Keep Sume job ids and keys outside it; re-read with jobs_status.
- Claude mcp_tool_listing pin: what a Sume tool list depends on
A pinned mcp_tool_listing holds the Sume tool list fetched for one session scope. Sume's list depends on mcp:read vs mcp:write and never pushes list changes.
- Claude Code 2.1.286 MCP Bearer redaction: Sume key cleanup
Claude Code 2.1.286 stopped MCP errors from printing a credential after Bearer. Update, then sweep old logs and rotate any Sume key that appeared there.
Written by Sume