Cursor Cloud Agent MCP Connector for Claude

Go to the **Cursor Dashboard → API Keys** section to create or manage your API key. The key starts with `crsr_` and can be either a user API key or a service account API key. Copy it and paste it into the API key field in the setup wizard. The key is sent as a Bearer token in the Authorization header for all API requests.

Cloud agents are autonomous coding agents that run in Cursor's managed environment. You give them a prompt describing what you want (e.g. 'Add a README file to this repo'), connect one or more repositories, and the agent works autonomously — reading code, making changes, and optionally creating pull requests. You can track run status, view results, and manage the agent lifecycle through the API.

Yes! When creating an agent, set the `autoCreatePR` flag to `true` and the agent will automatically create a pull request when it finishes its work. You can also filter agents by PR URL using the `pr_url` parameter in the `list_agents` tool to find agents associated with a specific PR.

Cursor cloud agents support multiple AI models including **gpt-4.1**, **claude-sonnet-4-20250514**, and **gemini-2.5-pro**. You can specify the model when creating an agent using the `model_id` parameter. If no model is specified, Cursor uses its default model. The model determines which AI processes your agent's prompts and generates code changes.

Use the `get_run` tool with the agent ID and run ID to get the full status including the current state (running, completed, cancelled), duration in milliseconds, the result output, and git branch information. You can also use `list_runs` to see all runs for an agent. To stop a long-running run, use `cancel_run` — this is a terminal operation that cannot be undone.

Yes. When creating a run with the `create_run` tool, you can pass an `mcp_servers_json` parameter containing a JSON array of MCP server configurations. This allows your cloud agent to access external tools and resources during the run, extending its capabilities beyond the default set of tools.

Yes, the Cursor Cloud Agents API enforces standard rate limiting. If you exceed the rate limit, the API returns an error and the engine surfaces the error message. For high-volume usage, consider spacing out requests and using pagination (limit and cursor parameters) when listing agents and runs. The `get_agent_usage` tool helps you monitor token consumption to stay within your plan's limits.