Where does the bug appear (feature/product)?
Cursor SDK
Describe the Bug
@cursorcursor/sdk v1.0.23, model: composer-2.5, React + TypeScript monorepo
Using the Cursor TypeScript SDK with local agents on a server, a task the IDE finishes in ~3m15s takes our headless run ~26 minutes — with the same result. The run log shows zero semantic searches and 122 edit calls to change 15 files. We’re trying to understand whether Fast Apply and semantic search reach local agents, and how to close the gap.
Do local SDK/CLI agents route edit through the Fast Apply (speculative-edits) apply model like the IDE, or does the planning model perform edits itself as sequential hunks? Is there any config, flag, or edit-tool choice (e.g. a batched / MultiEdit tool) to get one-shot / batched applies headlessly?
Is codebase_search / the semantic index available to local agents on a fresh server clone? Our runs show 0 semantic searches. If available, how do we build/enable the index for CI/server clones? If not, is it on the roadmap?
Is ~8× the edit round-trips (and no semantic search) the expected behavior for local agents today? Any recommended model, tool config, prompt, or workspace-scoping settings to reduce edit round-trips and speed up local headless runs?
Operating System
Linux
Version Information
@cursor/sdk v1.0.23
For AI issues: which model did you use?
Composer 2.5
For AI issues: add Request ID with privacy disabled
- request_id: 145c38fc-aea2-4e4a-bfbe-f85dc5735eab
- runId: run-322aec92-e1a8-4288-9444-a7007a09d1ef
- agentId: agent-8e493ab6-ea19-40a5-8bd2-a26363f3f692
- when (UTC): 2026-07-09 ~11:47:53 → 12:14, on the account tied to our CURSOR_API_KEY
Does this stop you from using Cursor
No - Cursor works, but with this issue