cursor

Cursor

Agent ID: cursor

Cursor's agent CLI. Speaks ACP natively via cursor agent acp - no bridge package needed.

Install

Install the Cursor app and ensure the cursor CLI is on your PATH. See the Cursor Agent CLI docs.

Sign in to the Cursor application to authenticate.

Quick Start

import { agents } from "@pivanov/agents-wire";

const result = await agents.ask(
  "cursor",
  "Explain this repo",
  {
    permission: "auto-allow",
  },
);

console.log(result.text);

Model Selection

cursor-agent acp does not accept a --model flag (verified via --help). options.model is forwarded as ACP modelPreference but is currently silently ignored by cursor-agent acp.

To list the models Cursor supports at runtime, call listAvailableModels():

import { definitionFor } from "@pivanov/agents-wire/catalog";

const def = definitionFor("cursor");
const models = (await def.listAvailableModels?.()) ?? def.models ?? [];

// models is IAgentModelOption[] - falls back to static list on any failure
console.log(models.map((m) => m.id));

Configure the active model via the Cursor application settings. options.effort is sent via ACP setSessionConfigOption (best-effort); Cursor's effort variants are typically baked into model IDs (kind: variant) - for example a model id might carry a -low, -high, or -xhigh suffix. Use listAvailableModels() to see the live IDs.

Capabilities

Feature	Supported
ask / stream / session	✅
askJson	✅
MCP stdio	✅
MCP http/sse	✅
Session listing (listSessions)	✅
Tool call interception	✅
Mode switching (setMode)	✅ (auto, max)

Cost Tracking

Cursor is subscription-based and does not report per-turn costUsd. cost.snapshot.totalUsd stays at 0, so maxCostUsd does not trigger. Monitor usage with cost.turnCount.

Auth Failure Detection

If the Cursor application is not signed in, the SDK catches the authentication-failure pattern in stderr and throws AgentUnauthenticatedError.