⚙Tool
Cursor
Launch and manage AI coding agents in Cursor from your workflows
Integrate Cursor's AI coding agents into Zelaxy workflows. Launch agents against a repository, send follow-up prompts, poll agent status, retrieve conversation history, and stop running agents — all from a single block. Use it to automate code-generation tasks, coordinate multi-step coding jobs, or build CI-style pipelines that invoke Cursor on demand.
Overview
| Property | Value |
|---|---|
| Type | cursor |
| Category | Tool — Developer |
| Auth | Basic (API Key — key sent as the username of an HTTP Basic credential) |
Operations
| Operation | Tool ID | Description |
|---|---|---|
| Launch Agent | cursor_launch_agent | Start a new AI coding agent against a repository with a prompt and optional model selection |
| Add Followup | cursor_add_followup | Send a follow-up prompt to a running or paused agent conversation |
| Get Agent | cursor_get_agent | Retrieve the status and details of a specific agent by ID |
| Get Conversation | cursor_get_conversation | Fetch the full conversation message history for an agent |
| List Agents | cursor_list_agents | List all agents with optional pagination |
| Stop Agent | cursor_stop_agent | Halt a running agent |
Configuration
| Setting | Type | Required | Description |
|---|---|---|---|
apiKey | string (secret) | Yes (all operations) | Cursor API key. Sent as the username of an HTTP Basic header — keep it in an env var and reference it as {{CURSOR_API_KEY}}. |
repository | string | Yes (Launch Agent) | Repository slug in org/repo format (e.g., acme/backend). Tells the agent which codebase to work in. |
ref | string | No (Launch Agent) | Branch or commit ref to check out (e.g., main, feature/my-branch). Defaults to the repo's default branch when omitted. |
promptText | string | Yes (Launch Agent, Add Followup) | The natural-language task description sent to the agent. For Launch Agent this is the initial prompt; for Add Followup it continues the existing conversation. |
model | string | No (Launch Agent) | Model to use for the agent (e.g., claude-4-5). Omit to use the Cursor default. |
agentId | string | Yes (Add Followup, Get Agent, Get Conversation, Stop Agent) | The unique agent ID returned by Launch Agent. Pipe it from an upstream block with {{launchBlock.id}}. |
limit | number | No (List Agents) | Maximum number of agents to return per page. |
cursor | string | No (List Agents) | Pagination cursor from a previous List Agents response (nextCursor) to fetch the next page. |
Outputs
| Field | Type | Description |
|---|---|---|
id | string | Agent ID (Launch Agent, Add Followup, Get Agent, Get Conversation, Stop Agent) |
url | string | Link to view the agent in the Cursor UI — https://cursor.com/agents?selectedBcId={id} (Launch Agent only) |
name | string | Agent name, if set (Get Agent) |
status | string | Current agent status, e.g., running, completed, stopped (Get Agent) |
source | json | Source repository information including repository slug and ref (Get Agent) |
target | json | Target branch information (Get Agent) |
summary | string | Agent summary produced after completion (Get Agent) |
createdAt | string | ISO timestamp of when the agent was created (Get Agent) |
messages | json | Array of conversation messages, each with role and content (Get Conversation) |
agents | json | Array of agent objects from the account (List Agents) |
nextCursor | string | Pagination cursor to pass back as cursor for the next page (List Agents) |
Example
[Starter] → [Cursor: Launch Agent] → [Cursor: Get Agent] → [Agent: summarize the result]A webhook trigger fires with a task description in {{starter.input.task}}. The first Cursor block launches an agent against {{REPO_SLUG}} on main using {{CURSOR_API_KEY}}, producing an agent id. A second Cursor block polls Get Agent with {{launchAgent.id}} to check the status. Once the workflow confirms completion, a downstream Agent block reads {{getAgent.summary}} and sends a Slack notification with the coding result.
Tips
- Pipe the agent ID — Launch Agent returns
idas its primary output; every downstream operation (Add Followup, Get Agent, Get Conversation, Stop Agent) requires it. Store it with a Variable block or pass it directly via{{launchAgent.id}}. - Poll with a loop — agents run asynchronously. Use a Loop block that calls Get Agent repeatedly and exits when
{{getAgent.status}}equalscompletedorstopped. - Auth format — Cursor uses HTTP Basic auth with the API key as the username and an empty password. You do not need to base64-encode it yourself; supply the raw key as
{{CURSOR_API_KEY}}and the tool handles encoding. - List Agents pagination — pass
{{listAgents.nextCursor}}back intocursoron the next iteration to walk through all results when you have many agents.