DOCSEARCHbyAlgolia
Submit Docs

Cursor Documentation

website·Indexed 13 days ago

https://cursor.com

Documentation for Cursor, an AI coding agent designed for codebase understanding, feature development, and bug fixing. It covers setup, the Agents Window, Plan and Debug modes, and tool integrations like Terminal, Browser, and Canvases. Includes detailed specifications for supported models from OpenAI, Anthropic, Google, and SpaceXAI, as well as guides on Cloud Agents, Model Context Protocol (MCP), and security configurations.

Tokens
113.9K
Snippets
358
Records
942
Agent score
50%

What's inside Cursor

  1. Overview of Cursor Background Agents (Cloud Agents)

    Background agents, also known as Cloud Agents, are coding agents designed for long-running tasks. Unlike local agents, they move the task to a cloud-based virtual machine (VM) where they can build features, fix bugs, and write tests from start to finish without interruption. Because they run remotely, users can close their local device and check the results later. Upon completion, these agents can automatically open pull requests.

  2. Overview of Cursor Plugins and Agent Plugins

    Plugins provide reusable tools and capabilities to the Cursor agent across desktop, web, and CLI environments. There are two primary types of plugins:

    1. Agent Plugins: Based on an open standard for packaging skills and MCP (Model Context Protocol) servers. Spec-conformant Agent Plugins load into Cursor without modification.
    2. Cursor Plugins: Support skills, MCP servers, as well as Cursor-specific rules, commands, and hooks.

    Plugins can be discovered via the official Cursor Marketplace or community sources like cursor.directory.

  3. Overview of Cursor Bugbot

    Bugbot is Cursor's automated pull request (PR) review tool. It analyzes PR diffs to identify bugs, security vulnerabilities, and code quality issues, providing inline comments with explanations and suggested fixes. It runs automatically on PR creation and updates, or can be triggered manually.

  4. Overview of Claude Fable 5 in Cursor

    Claude Fable 5 is a Mythos-class model designed for autonomous knowledge work and complex, long-running, and asynchronous coding tasks. It is currently one of the highest-performing models on CursorBench and is particularly effective at maintaining intent across long sessions with fewer user check-ins. A 'thinking' variant is available for problems requiring deeper reasoning.

  5. Overview of Cursor Blame AI attribution

    Cursor Blame extends traditional git blame to identify the origin of code lines. It categorizes authorship into three types:

    • Tab: Code generated or accepted from Cursor's Tab suggestions.
    • Agent: Code generated by an AI Agent (includes specific model attribution).
    • Human: Code written directly by developers.

    This feature is available exclusively on the Enterprise plan and requires a Git repository with Cursor-tracked changes.

  6. Overview of Cursor OpenTelemetry Export

    OpenTelemetry Export (Beta) allows Cursor Enterprise teams to stream usage data to a self-managed collector. It exports metrics (tokens, tool calls, estimated cost) and logs (API requests, errors, cloud agent events) server-side. This feature is available for the Enterprise plan and is configured in Team Settings > OpenTelemetry Export.

  7. Overview of Cloud Agent Environments

    Cloud agents run on isolated Ubuntu machines. A cloud agent environment consists of cloned repositories, installed dependencies, secrets, startup commands, and network access. Properly configured environments allow agents to run tests, query services, and reach APIs autonomously. Environments are managed via the Cloud Agents dashboard.

  8. Overview of Cursor Agent Browser capabilities

    The Cursor Agent can control a web browser to perform tasks such as testing applications, auditing accessibility, and converting designs to code. It has full access to console logs and network traffic for debugging and automation. The browser runs as a secure web view controlled via an MCP server extension and is integrated directly into the Cursor chat and layout.

  9. Overview of Cursor Cloud Agents

    Cloud Agents run in isolated VMs in the cloud rather than on a local machine. They provide a full development environment (cloned repos, dependencies, secrets, and network access) allowing them to build, test, and verify code changes. Key advantages include parallel execution, independence from local internet connectivity, support for multi-repo environments, and the ability to control a remote desktop and browser.

  10. Overview of Claude Sonnet 5 in Cursor

    Claude Sonnet 5 is Anthropic's medium-tier model that replaces Sonnet 4.6. It is designed to provide quality close to the Opus model while maintaining a lower per-token cost. Key features include support for 'thinking mode' for deep reasoning and a context window of up to 1 million tokens, making it suitable for complex coding tasks, planning, and debugging.

  11. Overview of Cursor Automations

    Cursor Automations run Cloud Agents in the background without manual input. They are triggered by schedules or events from external platforms such as GitHub, GitLab, Slack, webhooks, and Linear. Common use cases include reviewing pull requests, cleaning up feature flags, triaging bugs from Slack, and running security scans after CI completion.

  12. Overview of Cursor Worktrees

    Worktrees allow Cursor Agents to work in isolated Git checkouts. Each task receives its own set of files, dependencies, and changes, ensuring the main checkout remains untouched. This is particularly useful for running multiple agents on the same repository simultaneously without conflicts.