Critique Welcomes Cursor: Review Here, Fix There, Ship Together
You do not have to choose between deep PR review and the agent stack you already pay for. Wire Critique’s blueprint into Cursor Cloud Agents and stop treating “competitors” like mutually exclusive religions.
Critique × Cursor
Review depth. Your agent. One PR.
critique.sh
Critique welcomes Cursor.
Review depth from Critique. Execution on your Cursor Cloud Agent. One PR, one scoped blueprint — no wasted subscriptions and no pretending you have to pick a single winner.
Why compete when we can compose?
The AI tooling market loves cage matches. CodeRabbit vs Cursor Bugbot. Review bot vs coding agent. Every launch thread asks the same exhausted question: which one replaces the other?
That framing wastes money and attention. Most engineering orgs already pay for an editor agent—Cursor, Copilot, Codex, Claude Code. They also need merge confidence on GitHub: evidence-bound findings, policy-aware verdicts, and a record of what was checked before code lands on main.
Critique does not need Cursor to disappear. Cursor does not need Critique to pretend IDEs do not exist. What teams need is a clean handoff: **Critique produces the decision layer; Cursor executes the patch layer.**
Two jobs, one pull request
Mixing these jobs in a single agent is how you get confident-sounding fixes with weak evidence. Separating them is how you get speed without surrendering judgment.
| Question | Critique | Cursor Cloud Agent |
|---|---|---|
| Primary question | Should this merge as-is? | How do we fix what Critique flagged? |
| Success metric | Fewer false positives, clearer verdict, auditable findings. | Valid diff, passing checks, commits on the PR branch. |
| Billing | Critique credits for review depth. | Your Cursor plan / Cloud Agents usage. |
| Where it runs | GitHub App checks, sandbox review, dashboard artifact. | Cursor-hosted VM against your repo + PR. |
| When to use | Every PR that needs an independent review pass. | After you trust the findings and want execution. |
What we shipped
The integration is deliberately boring in the best way: encrypted Cursor API key in Settings, **Queue Cursor agent** on the review run, QStash worker calls the Cloud Agents API with `prUrl` and `workOnCurrentBranch`, you watch the run in Cursor. Export-only path still exists—download `critique.cursor_agent_handoff` JSON if you want SDK or CLI control.
- Repository + PR number + head SHA
- Scoped allowed write files from the review artifact
- Validation commands (lint, test, typecheck)
- Stop conditions and operator instructions
- Full Remedy-compatible blueprint embedded in the handoff
- Cursor API key (Dashboard → Integrations)
- Cloud Agent execution and model choice
- Commits pushed to the PR branch
- Cursor usage billing
Remedy vs Cursor BYOA: pick the contract you want
Critique already runs **Remedy**—managed fix execution in isolated E2B sandboxes with guarded pushback. That path is right when you want Critique to own the runtime, the validation loop, and the operational envelope.
**Cursor BYOA** is the alternative when you already pay for Cursor and want execution there. Same blueprint. Different executor. Critique credits focus on review; Cursor usage focuses on the fix. No double-paying for the same cognitive work.
Critique-managed execution. Best when you want one vendor to run the sandbox, validation, and push.
POST /api/review-runs/{id}/remedyYour Cursor Cloud Agent. Best when your team lives in Cursor and wants fixes on your existing agent subscription.
POST /api/review-runs/{id}/cursor-agentWhy this is better than “just use Bugbot”
Cursor Bugbot is a strong product when Cursor is already the center of gravity. Critique is for teams that want **review independence**: multi-model specialist passes, transparent credit economics, repo chat grounded in review artifacts, and a fix path that does not require standardizing on one editor.
Welcoming Cursor as an execution surface does not make Critique a Bugbot clone. It makes Critique honest about the stack map: **we review; you may fix with whatever agent you already bought.**
The developer experience we are optimizing for
Picture a staff engineer’s afternoon. A PR opens. Critique runs on sync. The check fails with three findings—one security, one test gap, one API footgun. The engineer agrees with two, disagrees with one, replies on the thread. For the two real issues, they hit **Queue Cursor agent** with a one-line operator note: “only touch `auth/` and matching tests.”
They open the agent in Cursor, watch the cloud run, see validation pass, refresh the PR. They never pasted a megabyte prompt. They never re-explained repo layout. Critique stayed the source of truth for **what** to fix; Cursor stayed the source of truth for **how** to edit.
How to turn it on
- 11. Migration applied?Your Critique deployment needs the cursor_agent_run tables (May 2026 migration).
- 22. Secrets configured?CRITIQUE_SECRETS_ENCRYPTION_KEY must be set so API keys can be stored.
- 33. Cursor key?Settings → Cursor API key → paste from cursor.com/dashboard/integrations.
- 44. Review finished?Open a completed review run → Cursor agent section → Queue Cursor agent.
- 55. Optional model override?Set CRITIQUE_CURSOR_AGENT_MODEL_ID on the server if you want a fixed Cloud Agents model id.
Optional: `GET /api/review-runs/{reviewRunId}/byoa/cursor` returns the handoff JSON for your own automations—CI that launches agents, internal bots, or a future `@cursor/sdk` script.
What we will not pretend
This integration does not make Cursor runs invisible inside Critique. You still open Cursor to see tool calls, branches, and failure modes. Critique records agent id, run id, status, and a result summary for the dashboard.
We also will not claim Critique controls Cursor pricing, model availability, or org policies. Your key, your account, your rate limits. Critique’s job ends at a well-formed handoff and an honest status line.
The bigger bet
AI code tools will keep specializing. Review will get deeper—more evidence, more policy, more graph context. Execution will get faster—better agents, better sandboxes, better IDE surfaces. The winners in 2026 are not monoliths that do everything adequately. They are **composable layers** teams can route work through without renegotiating their entire stack every quarter.
Critique welcoming Cursor is that bet in product form. Use us for the merge decision. Use Cursor for the patch. Use Remedy when you want us to run the fix. Use Fix Prompt when you want a paste-friendly escape hatch. Stop forcing one logo to swallow a workflow it was never designed to carry alone.
Wire your first Cursor handoff
Install Critique on your repo, complete a review, connect your Cursor API key, and queue an agent on the PR you already have open.
Open Settings