cursor2api

中文说明

Use the models of your Cursor account through the Anthropic Messages API and the OpenAI Chat Completions API.

• POST /v1/messages and POST /v1/chat/completions, streaming or buffered
• every model the signed-in account can use, listed by GET /v1/models
• tools, images, PDFs, thinking/reasoning, usage
• authorise with a Cursor API key or a browser login

Install

git clone <your fork>
cd cursor2api
pip install -r requirements.txt

Python 3.9+ and the h2 package.

Authorise

Either set a key from cursor.com/dashboard:

export CURSOR_API_KEY=crsr_...

or authorise in the browser:

python -m cursor2api login

serve also does it by itself: with no credentials it prints an authorisation URL and waits for you to approve it. A server started in the background instead answers GET /login with a URL and picks up the approval on its own:

curl -s localhost:8787/login

Tokens are kept in ~/.config/cursor2api/credentials.json (0600). status shows what is in use, logout removes it.

Use

python -m cursor2api serve            # http://127.0.0.1:8787

Anthropic clients, Claude Code included:

ANTHROPIC_BASE_URL=http://127.0.0.1:8787 ANTHROPIC_API_KEY=sk-local claude

curl -s localhost:8787/v1/messages -H 'content-type: application/json' -d '{
  "model": "claude-sonnet-4-5", "max_tokens": 256,
  "messages": [{"role": "user", "content": "hello"}]}'

OpenAI clients, same port:

OPENAI_BASE_URL=http://127.0.0.1:8787/v1 OPENAI_API_KEY=sk-local

curl -s localhost:8787/v1/chat/completions -H 'content-type: application/json' -d '{
  "model": "gpt-5.6-sol", "stream": true,
  "messages": [{"role": "user", "content": "hello"}]}'

Routes: POST /v1/messages, POST /v1/messages/count_tokens, POST /v1/chat/completions, GET /v1/models, GET /v1/models/{id}, GET /login, GET /health.

Common settings: PORT, BIND, API_KEY (require an x-api-key from local clients), DEFAULT_MODEL.

Models

GET /v1/models returns the account's own catalog. Any of these spellings works as model:

form	example
base model	claude-fable-5, gpt-5.6-sol, gemini-3.1-pro, kimi-k3
Cursor variant	claude-fable-5-thinking-xhigh, composer-2.5-fast
alias	fable, sonnet-latest, opus, codex
explicit parameters	claude-sonnet-5[thinking=false,effort=max]
another vendor's id	claude-3-5-sonnet-20241022, gpt-4o (mapped to the nearest model)

Unknown ids fall back to DEFAULT_MODEL.

Notes

• temperature, top_p, top_k, cache_control, n, seed and response_format have no equivalent upstream and are ignored; stop_sequences, max_tokens and tool_choice are approximated locally.
• Thinking text is Cursor's summary and its Anthropic signature is always "".
• Web search is Cursor's own server-side tool, reported as server_tool_use + web_search_tool_result with titles and urls only.
• Cursor's agent system prompt is always present and inflates input_tokens.
• Rate limits come from the Cursor account and surface as 429 with retry-after.
• A model the account has not enabled (some require accepting a data retention policy in the Cursor dashboard) answers 403 permission_error.

Protocol details and field numbers: docs/protocol.md.

Tests

Start the server, then:

python tests/test_api.py
python tests/test_openai.py

Disclaimer

This project is an independent, unofficial and experimental tool. It is not affiliated with, endorsed by or supported by Cursor (Anysphere), Anthropic or OpenAI, and the names are used only to describe wire formats.

It talks to a private, undocumented and version-gated protocol, so it can stop working at any time. Using it may conflict with Cursor's terms of service and with the terms of the model providers behind it; you are responsible for how you use it and for your own account. No warranty of any kind: see LICENSE.

License

MIT.

Repository topics

cursor cursor-ai cursor-api anthropic-api claude openai-api openai-compatible
anthropic-compatible llm-proxy api-proxy reverse-engineering protobuf connect-rpc
python sse streaming function-calling claude-code