# Cursor Agent SDK (slim)

> Cursor's TypeScript framework for defining an agent as a folder of instructions, typed tools, skills, MCP connections and channels, testing it with evals, and running it on your own server or on Cursor's hosting.

- Full: https://www.anchorterminal.com/tools/cursor-agent-sdk.md (~8,750 tokens) · this version ~1,730 tokens · JSON https://www.anchorterminal.com/tools/cursor-agent-sdk.json · canonical https://www.anchorterminal.com/tools/cursor-agent-sdk
- Index: https://www.anchorterminal.com/llms.txt · API: https://www.anchorterminal.com/api/v1/index.json · MCP: https://www.anchorterminal.com/mcp · Updated: 2026-10-27 · pricing https://www.anchorterminal.com/tools/cursor-agent-sdk#pricing · reviews https://www.anchorterminal.com/tools/cursor-agent-sdk#reviews
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)

**B · 63.6/100 · rank #443 of 1217 · #12 in Agent frameworks & SDKs · not agent-ready · confidence medium**


Assessment: An agent-as-code framework with typed tools, declared read and write effects that drive retries and dry runs, an approval gate per tool and evals that assert on tool calls. It's closed source, every model turn runs through Cursor's backend on a Cursor plan, local sandboxing is off by default, and the changelog has two entries.

## Facts

- Kind: Agent framework · vendor: Cursor · category: Agent frameworks & SDKs · legal entity: Anysphere, Inc. · provenance 91/100
- Packages: npm `@grok/agent-sdk`
- Auth: OAuth or key · pricing: Your plan · x402: no · licence: Proprietary. The package's LICENSE.md reads "© Anysphere Inc. All rights reserved", with use subject to Cursor's Terms of Service (updated 3 September 2026). No source repository is published
- Surface listed: The npm package @grok/agent-sdk 0.3.0 and its `agent-sdk` CLI, read from the docs and the published tarball on 27 October 2026. Not installed or run
- Language: TypeScript on Node 22.13 or newer, never Bun. Tools take Zod or plain JSON Schema inputs
- Project: bot/ holds agent.ts, instructions.md, tools/, skills/, mcp-connections/, host-connections/, subagents/, extensions/, channels/, hooks/, schedules/ and reminders/, with evals/ beside it
- Models: Cursor models through the Cursor backend. Default grok-4.5 with effort high and Fast on. Third-party models at list price on Premium and above
- Runs on: `agent-sdk serve` on your own host (`127.0.0.1:3000` by default), or `agent-sdk deploy` of a pushed Git ref to Cursor-managed hosting with an egress allowlist and named secrets
- MCP client: Remote HTTP, local stdio, the Cursor account's dashboard connectors, and peer agents by slug, with allow and deny lists and a read-only filter
- Multi-agent: Subagents started by the `task` tool, and peer agents reached through their `ask`, `check` and `call_tool` tools
- Durable state: Sessions, a key-value store and files survive restarts and redeploys. Follow-ups on a pull request or Slack thread resume the same session
- Human approval: `needsApproval` parks a tool call until a person approves or denies it. Dry-run sessions run reads and stub writes
- Channels: GitHub, GitLab and Bitbucket events, Slack mentions and DMs, PagerDuty, signed webhooks from any provider, and cron schedules
- Telemetry: OpenTelemetry export only when an OTEL endpoint is set, with prompt text and tool payloads behind separate flags. A daily npm version check, off with AGENT_SERVE_NO_UPDATE_CHECK
- Hosting limits: Where Cursor caps a team, 20 deployments and 100 deploys in 24 hours by default. A deploy can answer 503 with Retry-After
- Package names: @grok/agent-sdk in the docs. @cursor/bdk and the early-alpha @cursor/july publish the same 0.3.0
- Not to confuse with: Cursor CLI (cursor-cli), the terminal coding agent, and the Cursor SDK (@cursor/sdk), the harness this package runs on
- Prices: Grok 4.5 Fast (the default model) input, beyond the weekly pool $4 per 1M tokens; Grok 4.5 Fast (the default model) output, beyond the weekly pool $18 per 1M tokens; Grok 4.5 input, Fast off, beyond the weekly pool $2 per 1M tokens; Grok 4.5 output, Fast off, beyond the weekly pool $6 per 1M tokens; Premium plan (first plan with Cloud Agents and third-party models) $30 per month (plan)

- Scores: Reliability 45, Performance pending, Schema & documentation 82, Agent ergonomics 80, Security & auth 68, Payments & pricing 30, Task success pending, Maintenance & community 74, Transparency & trust 70 · total over the 7 assessed categories
- Why: Reliability, Local-package reading, since the owner runs the framework, with Cursor's hosting as an option. · Schema & documentation, Framework reading. · Agent ergonomics, Framework reading, scored on what it takes to run a tool-calling agent with MCP. · Security & auth, Framework reading. · Payments & pricing, Framework reading of the published rubric. · Maintenance & community, 0.3.0 on npm on 25 October 2026 (30). · Transparency & trust, Closed source.
- Sources: 37, open questions: 8, both in the full twin
- Capabilities: agent.framework, agent.mcp-client, agent.durable, agent.multi-agent, obs.evals
- JSON: https://www.anchorterminal.com/api/v1/tools/cursor-agent-sdk.json
- Verify (for the vendor): the badge `https://www.anchorterminal.com/badges/cursor-agent-sdk.svg` or a link to https://www.anchorterminal.com/tools/cursor-agent-sdk from a page on cursor.com or one of its subdomains, then `POST https://www.anchorterminal.com/api/v1/verify` `{"slug", "url"}` or `verify_listing` at /mcp; re-checked weekly, no effect on the grade. Snippets in the full twin.

## Before you call it

1. Install `@grok/agent-sdk`, not `@cursor/agent-sdk` from the launch post. That name answers 404 on npm
2. Set `local.sandbox` before a turn reads webhook payloads, pull request diffs or chat. Without it shell commands auto-approve with the host's credentials
3. Give every tool an `effect`. A tool without one counts as a write, runs once and is stubbed in dry runs
4. Start `agent-sdk serve` with `--bearer-token` on any shared host. `--allow-anonymous` admits every caller as one principal
5. Delete bot/hooks/memory.ts when sessions serve different people. The journal shares each turn's message and result with later turns

## Connect

```bash
npm install --global @grok/agent-sdk
```

## Similar tools

| Tool | Grade | Score | Shared capabilities | Slim |
| --- | --- | --- | --- | --- |
| OpenAI Agents SDK | AA | 86.2 | agent.framework, agent.multi-agent, agent.durable, agent.mcp-client | https://www.anchorterminal.com/tools/openai-agents-sdk.min.md |
| Microsoft Agent Framework | A | 82.2 | agent.framework, agent.multi-agent, agent.durable, agent.mcp-client | https://www.anchorterminal.com/tools/microsoft-agent-framework.min.md |
| Pydantic AI | A | 81.7 | agent.framework, agent.multi-agent, agent.durable, agent.mcp-client | https://www.anchorterminal.com/tools/pydantic-ai.min.md |
| Docker Agent | BB | 76.5 | agent.framework, agent.multi-agent, agent.mcp-client, agent.durable | https://www.anchorterminal.com/tools/docker-agent.min.md |
| Agent Development Kit (ADK) | BB | 73.6 | agent.framework, agent.multi-agent, agent.durable, agent.mcp-client | https://www.anchorterminal.com/tools/google-adk.min.md |


## Panel reviews (0, desk reviews from public material, no calls made)