Answer: Agent SDK automatically retries a transient read or idempotent call up to 3 times before the model sees the error. It retries an MCP tool when the tool's annotations mark it read-only, or when the tool sets `idempotentHint: true` together with an explicit `destructiveHint`. Status: Confirmed Entry: tool/cursor/agents-and-automation/retry-transient-call Revision: 1 Digest: sha256:07107391969bd4dd (changes whenever the content changes) Timeline: Created: 2026-11-13T03:09:05Z Last changed: 2026-11-13T03:09:05Z Last verified: 2026-11-20T12:34:19Z Next check: 2026-11-21T12:34:19Z Check interval: Daily Expected change: No fixed schedule Data: Field | Value | Unit | Source | As of retry attempts | 3 | times | [1] | 2026-11-20 read-only annotation condition | read-only | | [1] | 2026-11-20 idempotent hint condition | idempotentHint: true | | [1] | 2026-11-20 writes and undeclared-effect tools | only once | | [1] | 2026-11-20 error after final failure | the last attempt's message | | [1] | 2026-11-20 Sources: [1] Anysphere. "cursor.com/docs/agent-sdk/reference/tools.md". https://cursor.com/docs/agent-sdk/reference/tools.md Role: primary. Accessed: 2026-11-20T12:34:19Z Quote: "tries it up to 3 times before the model sees the error" (Supports: retry attempts) Quote: "Agent SDK retries an MCP tool when the tool's annotations mark it read-only" (Supports: read-only annotation condition) Quote: "sets `idempotentHint: true` together with an explicit `destructiveHint`" (Supports: idempotent hint condition) Quote: "It runs writes, and tools that don't declare an effect, only once." (Supports: writes and undeclared-effect tools) Quote: "If the call still fails, the tool error is the last attempt's message." (Supports: error after final failure) Snapshot SHA-256: 478596a021cb5a0713ad781c7e9ed2aab5f83d2769eb6e2f6bb795941c038cfb (captured 2026-11-13T03:09:05Z) Verification: Method: Automated extraction from [1] with deepseek-flash (JSON output, thinking disabled). Every value was checked by code to appear verbatim in a quoted passage of [1]. On this check the text of [1] was unchanged since 2026-11-13T03:09:05Z, so the values verified then were kept. Cross-check: not-available at 2026-11-20T12:34:19Z. No independent second source is configured for this entry Last successful check: 2026-11-20T12:34:19Z Last attempt: 2026-11-20T12:34:19Z Notes and limitations: Extracted automatically from a single official source (section "Retry a transient read or idempotent call"). Related: https://latestn.com/tool/cursor/agents-and-automation/review-pull-request-triggers How do I make a Cursor GitHub agent review pull requests only when opened or ready for review? https://latestn.com/tool/cursor/agents-and-automation/reviewer-assignment How does Cursor PR Routing & Approval assign reviewers to pull requests? https://latestn.com/tool/cursor/agents-and-automation/risk-based-approval How does risk-based approval work in Cursor PR Routing & Approval? https://latestn.com/tool/cursor/agents-and-automation/read-agent-metadata How do I read agent metadata from inside a Cursor Cloud Agent VM? https://latestn.com/tool/cursor/agents-and-automation/remote-control-handoff How do I hand off a local Cursor session to my phone using Remote Control? https://latestn.com/tool/cursor/agents-and-automation/restart-from-plan How do I start over from a plan in Cursor? https://latestn.com/tool/ All entries in AI tools Change log: Time | Event | Change | Source 2026-11-13T03:09:05Z | added | Entry created | — Changes for this entry as a feed: https://latestn.com/tool/cursor/agents-and-automation/retry-transient-call/latest Challenge this entry: Open challenges: None How: If a value here is wrong or out of date, send what is wrong and a link to evidence. Challenges are queued and checked against the evidence; they do not change the entry by themselves. If the entry changes as a result, the change log says so. Form: https://latestn.com/dispute/tool/cursor/agents-and-automation/retry-transient-call POST: https://latestn.com/dispute/tool/cursor/agents-and-automation/retry-transient-call with claim, evidence_url, and optionally field and contact (form-encoded or JSON) GET: https://latestn.com/dispute/tool/cursor/agents-and-automation/retry-transient-call?claim={what is wrong}&evidence_url={link} How to cite: LatestN. (2026). How do I retry a transient read or idempotent call in a Cursor Agent SDK tool? Revision 1, last verified 2026-11-20T12:34:19Z. https://latestn.com/tool/cursor/agents-and-automation/retry-transient-call (accessed 2026-11-21T10:07:16Z). @misc{LatestN_tool_cursor_agents_and_automation_retry_transient_call_r1, title = {How do I retry a transient read or idempotent call in a Cursor Agent SDK tool?}, author = {{LatestN}}, year = {2026}, howpublished = {\url{https://latestn.com/tool/cursor/agents-and-automation/retry-transient-call}}, note = {Revision 1; digest sha256:07107391969bd4dd; last verified 2026-11-20T12:34:19Z; accessed 2026-11-21T10:07:16Z} } Reuse: License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) Attribution: Copying and adapting are allowed, including commercially, if you credit LatestN and link to the entry you used. Each entry ends with a ready-made citation. URL: https://latestn.com/tool/cursor/agents-and-automation/retry-transient-call