Skip to main content

Cursor

Cursor is an AI-native code editor built on a VS Code fork, with agentic coding capabilities in its GUI (Composer, agent mode).

AgentXchain does not currently ship a native Cursor connector. Cursor's strongest AI features are GUI-only. The cursor CLI command opens files and the editor, but it does not expose a documented headless mode for programmatic agent interaction. The local_cli adapter cannot spawn Cursor as a governed subprocess today.

The closest supported path is to use Cursor as your editing environment while connecting a separate CLI agent as the governed runtime.

Read the Authority Model before you configure the terminal-side runtime. Cursor is the editor, not the governed runtime.

Recommended setup: Cursor as editor + CLI agent runtime​

Edit your code in Cursor. Run AgentXchain in Cursor's integrated terminal with a proven CLI agent handling the governed turns.

Configuration (Claude Code as runtime)​

{
  "runtimes": {
    "claude-dev": {
      "type": "local_cli",
      "command": ["claude", "--print", "--dangerously-skip-permissions", "--bare"],
      "cwd": ".",
      "prompt_transport": "stdin"
    }
  },
  "roles": {
    "dev": {
      "runtime": "claude-dev",
      "mandate": "Implement features following project conventions",
      "write_authority": "authoritative"
    }
  }
}

Configuration (OpenAI Codex CLI as runtime)​

{
  "runtimes": {
    "codex-dev": {
      "type": "local_cli",
      "command": ["codex", "exec", "--dangerously-bypass-approvals-and-sandbox", "{prompt}"],
      "cwd": ".",
      "prompt_transport": "argv"
    }
  },
  "roles": {
    "dev": {
      "runtime": "codex-dev",
      "mandate": "Implement features following project conventions",
      "write_authority": "authoritative"
    }
  }
}

Authority mapping​

  • write_authority: the governed role should usually be authoritative for local CLI implementation turns.
  • Runtime type: local_cli on the separate Claude Code or Codex runtime, not on Cursor itself.
  • Downstream CLI authority: the terminal-side tool still needs its own unattended-write flag (--dangerously-skip-permissions for Claude Code, --dangerously-bypass-approvals-and-sandbox for Codex).

Minimal working example​

agentxchain init --governed --template web-app --goal "Build a bookmarks manager" --dir my-project -y
cd my-project
# Edit agentxchain.json to use one of the CLI runtime configs above.
agentxchain doctor
agentxchain connector check
agentxchain connector validate <runtime_id>   # use claude-dev or codex-dev from your config
agentxchain run

Or use the guided interactive path (prompts for template, name, goal, and folder):

agentxchain init --governed

Then update agentxchain.json with the CLI runtime you chose, run agentxchain doctor, agentxchain connector check, and agentxchain connector validate <runtime_id>, then start agentxchain run.

Using Cursor's built-in AI alongside AgentXchain​

Cursor's Composer, Chat, and agent mode work inside the editor while AgentXchain governs the structured delivery process in the terminal. They are complementary:

  • Cursor AI — ad-hoc coding assistance within the editor (autocomplete, refactoring, chat)
  • AgentXchain — governed multi-agent turns, role enforcement, artifact tracking, phase gates
  • Authority model — still defined by the separate governed runtime in the terminal, not by Cursor itself

Future: if Cursor adds a headless CLI mode​

If Cursor ships a documented headless CLI mode for agentic coding in the future, the generic local_cli adapter would support it with a config like:

{
  "runtimes": {
    "cursor-dev": {
      "type": "local_cli",
      "command": ["cursor", "<future-cli-flag>"],
      "cwd": ".",
      "prompt_transport": "stdin"
    }
  }
}

This is speculative and not a working config today. Check Cursor's documentation for the latest CLI capabilities.

Gotchas​

  • No headless CLI mode: Cursor's AI capabilities (Composer, agent mode, Chat) are GUI-only. You cannot spawn Cursor as a subprocess for governed turns.
  • CLI opens the editor: cursor on the command line opens the Cursor GUI or a file in the editor — it does not start a headless coding session.
  • Terminal works fine: Cursor's integrated terminal runs agentxchain commands normally — the limitation is only about using Cursor as the agent runtime.
  • Model routing: Cursor uses its own model routing layer. The models available in Cursor are separate from the models you configure in your AgentXchain runtime.