We’re introducing the Cursor SDK so you can build agents with the same runtime, harness, and models that power Cursor. Run agents from CI/CD pipelines, create automations for end-to-end workflows, or embed agents directly inside your products.

With the Cursor SDK, you can run agents locally or deploy them in our cloud.

We've open-sourced a few starter projects for you to build on: a coding agent CLI, a prototyping tool, and an agent-powered kanban board. Use Cursor to customize them for your use case:

Customers like Rippling, Notion, C3 AI, and Faire are using the Cursor SDK to build custom background agents, take bugs from ticket to merge-ready PR, and maintain self-healing codebases. Learn more:

anthropic: here's claude code cursor: here's an sdk to build on top of that developers: hold on let me quit my job real quick

for real

would be pretty cool if paying users could use the SDK with their sub, regardless this is good.

@cursor_ai just keeps raising the bar. SDK looks incredible.

AI news these days

the SDK shipping is the moment cursor stops being an IDE and becomes a platform.. agents in CI/CD means you can run cursor without ever opening cursor interesting tho that this lands the same week google dropped $40B on anthropic for TPU capacity. every coding tool needs hyperscaler compute now. cursor ships the SDK, anthropic ships the model, google sends the bill

we're so back

this is the future. gonna switch my projects to this now!

wrote the vibe coder translation of this announcement. tldr: cursor stopped being an editor today.

Don't let those sweet sweet traces go to waste

Agents running in CI/CD pipelines generate outputs and artifacts continuously. The harness manages execution. The storage layer underneath determines whether those outputs are persistent and auditable

Smart tbh -- one of the big things this does is get Cursor's tools & harness in the training data for the next models from all the frontier providers. Models will become even better in Cursor.

Looks nice

It’s crazy to think that Cursor started just with AI autocomplete, and now we’re at this point

the real unlock here is building agents with the *same* stuff cursor uses consistency is king for dev tools especially AI ones

this is the right move. letting people run cursor agents outside the editor changes what the product actually is.

It's a good agent. No, wait: It's the best agent! @cursor_ai Agent CLI achieved the highest score in all of @WolfBenchAI across 23 models and 5 agents in more than 300 runs. And it did so with GPT-5.5 at only the high reasoning/thinking level - it didn't even need xhigh!

cursor just handed devs the keys to their own jailbreak finally a sdk that doesn't feel like enterprise ransomware

Nice one @cursor_ai Actually open-sourcing the real harness and sandbox you use in Composer is smart. Excited to play with the SDK.Good stuff.

Giving away the engine that makes Cursor special is a massive risk. Will this just turn every mediocre IDE into a clone?

now that’s awesome

Actually this could be pretty sweet, especially if your company has private data you want it to have access to

I see it as the next generation of AI software ... Here is my view -

This is huge 🔥 Finally, the full Cursor runtime in our own apps and pipelines. The ability to embed these agents directly into products + run them in CI/CD is going to change how we build AI-native software. Already thinking about agentic workflows that can take a ticket → write code → test → deploy. When can we get our hands on the SDK? 👀

cursor just productized the harness. the model is a config string. the agent is the import. this is the same shift openai admitted on monday and the same shape perplexity shipped with their qwen post-train this week. three companies in seven days saying out loud that the model layer is commoditizing and the harness layer is the product. anthropic is next.

Been building agents that break in CI because the context is so different from development. This could actually solve that.

hit this constraint. Agents only work in the ide context, blind to logs, failures, actual deployments. Embed them in ci/cd and they can finally see the full system. Whole different tier.

Totally incredible, can’t wait to play around with this! 🔥

Amazing. OpenAI started the Agents SDK initiative way earlier. It's smooth and have best in-class DX. If same harness + extra features Cursor provides then I'll probably think of switching my production-grade ai-native workflows.

If cursor’s entire business model compresses to the agent harness, wouldn’t an SDK ease the reverse engineering of it?

Cursor just opened the back door to agents. New frontiers are coming.

Build agents with the same engine that powers Cursor itself? They just opened the whole thing up.

CURSOR TEAM BEEN COOKING LATELY

The interesting claim is 'same runtime and harness' — not just the models. The messy edge cases Cursor has already handled carry over instead of you rediscovering them in your own pipeline. Worth watching how well that holds once it's under real load.