Cursor (Legacy) Block

The Cursor (Legacy) block integrates the Cursor AI coding assistant into your workflows. Use it to launch coding agents against a repository, send follow-up instructions, check agent status, retrieve conversation history, list agents, and stop agents.

Overview

When to Use

• Kick off an AI coding agent on a specific repository and branch with a natural-language task prompt
• Send a follow-up prompt to an already-running agent to refine or extend its work
• Poll an agent's status to gate downstream steps in a workflow until the agent finishes
• Pull back the full conversation history for an agent for logging or summarisation
• List all active agents across your Cursor workspace to find one to act on
• Stop a running agent that is no longer needed to save compute

Configuration

Operation

Required dropdown that selects which Cursor API action the block performs. Every other field appears only when its operation is selected.

Repository

The target repository slug in org/repo format. Shown for Launch Agent only. Example: acme/backend.

Branch/Ref

The branch or commit ref the agent should work against. Shown for Launch Agent only. Defaults to the repository's default branch when omitted. Example: main.

Prompt

The natural-language task description for the agent. Shown for Launch Agent and Add Followup. For a launch this is the initial instruction; for a follow-up it extends the existing conversation.

Agent ID

The identifier of an existing Cursor agent. Shown for Add Followup, Get Agent, Get Conversation, and Stop Agent. Use {{cursor_1.agentId}} to pass the ID from a previous Launch Agent step.

Model

The model the launched agent should use. Shown for Launch Agent only. Optional; omit to use the Cursor default. Example: claude-4-5.

Inputs & Outputs

• Inputs:

  • operation (string) — Operation to perform (one of the tool IDs above)
  • repository (string) — Repository slug (e.g. org/repo)
  • ref (string) — Branch or ref
  • promptText (string) — Agent prompt
  • agentId (string) — Agent ID
  • model (string) — Model name

• Outputs:

  • agentId (string) — Agent ID (returned by Launch Agent; pass to subsequent steps)
  • status (string) — Agent status (returned by Get Agent)
  • conversation (json) — Conversation history (returned by Get Conversation)

Tools

• Cursor Launch Agent (cursor_launch_agent) — POSTs to https://api.cursor.com/v0/agents to create a new AI coding agent for a repository, branch/ref, prompt, and optional model. Returns the new agent id and a url to view the agent in Cursor.
• Cursor Add Followup (cursor_add_followup) — POSTs a follow-up prompt to an existing agent's conversation (/agents/{agentId}/followup). Returns the follow-up message id.
• Cursor Get Agent (cursor_get_agent) — GETs the status and metadata of a single agent (/agents/{agentId}). Returns id, name, status, source, target, summary, and createdAt.
• Cursor Get Conversation (cursor_get_conversation) — GETs the full conversation history for an agent (/agents/{agentId}/conversation). Returns the conversation id and an array of messages.
• Cursor List Agents (cursor_list_agents) — GETs a paginated list of all agents. Accepts optional limit and cursor pagination params. Returns an agents array and a nextCursor for the next page.
• Cursor Stop Agent (cursor_stop_agent) — POSTs to /agents/{agentId}/stop to halt a running agent. Returns the stopped agent id.

All tools authenticate using HTTP Basic auth with your Cursor API key ({{CURSOR_API_KEY}}).

YAML Example

cursor_1:
  type: cursor
  name: "Cursor (Legacy)"
  inputs:
    operation: "cursor_launch_agent"
    repository: "acme/backend"
    ref: "main"
    promptText: "{{trigger.userRequest}}"
    model: "claude-4-5"
  connections:
    outgoing:
      - target: next-block-id