Cursor Automations + SDK

Turn Cursor Agents into functional members of your team.

Turn any signal into a coordinated team of agents. Choose a proven playbook with fixed actions, models, gates, and outputs, then watch Cursor run it on its own computer.

Explore prebuilt automationsInteractive · deterministic · no live systems touched

Start with your stack

Pick the tools your team already uses. Cursor separates native triggers, webhooks, and MCP access so the wiring stays honest.

Compose the agent team

Give each agent a job, model, filter, and handoff. Triage with Opus, edit with Composer, review with Codex.

Watch every handoff

Run the exact automation you built. The orb follows each agent as it gathers evidence, edits, tests, and ships.

Open the evidence

Inspect the run log, recorded test, pull request, ticket update, and team-channel summary your workflow produced.

Automations

Start from a proven playbook

Pick the job your team needs done. Every trigger, agent action, evidence gate, model, and output is already wired together.

SDK

Own the orchestration

Inspect the skills, MCP connections, models, and guardrails behind each playbook, then deploy the same pattern inside your controls.

What your team deploys

Triggers, scoped tools, agent instructions, and review gates.

The demo is scripted for safety. The scaffolding is concrete: Automations start Cloud Agent runs, prompts define each handoff, and repo-local skills make the behavior reviewable.

Runtime architecture

6 native · 11 adaptable triggers

Any event may start a discrete run

→

Model-specific Cloud Agents

Filter · handoff · edit · review · test

→

26 demo tools · 4 real skills

MCPs, code, browser, artifacts, and humans

Powerful does not mean unbounded

Every agent runs inside controls your team owns.

Isolated Cloud Agent

Each run starts in its own configured VM or approved self-hosted worker.

Scoped tools and secrets

Triggers do not silently grant write access. Tools, MCPs, and runtime secrets stay explicit.

Filters before writes

Evidence gates can stop noisy events before an editor, PR, ticket, or message step runs.

Human review remains

Agents can propose and verify changes. The final merge stays with your reviewers.

Controlled network access

Teams can constrain egress, models, computer use, MCPs, and long-running agents.

Evidence with every run

Tool calls, tests, screenshots, recordings, and outputs remain attached to the run.

Find the playbook closest to work your team still does by hand.

Run it end to end, inspect every artifact, and use the fixed action chain as the starting point for your production automation.

Open the playbook gallery