Cursor with Plan2API: custom models through your own key

Settings at a glance

Every value in the table below goes into Cursor Settings > Models.

Set up Cursor with Plan2API

1. Sign in at plan2api.com/connect with a Google account, connect a Claude Pro or Max, ChatGPT, Google AI Pro or Ultra, or Kimi Code plan, and create a p2a_ API key.
2. Test your new key from a terminal with the curl command below to confirm your subscription is connected and responding before you change any Cursor settings.
3. Open Cursor Settings > Models in your editor.
4. Paste your p2a_ key into the OpenAI API Key field and save it.
5. Turn on Override OpenAI Base URL and enter https://api.plan2api.com/v1.
6. Add each model you want to use by its exact Plan2API model id, such as claude/claude-opus-5-5, openai/gpt-6.1-sol, gemini/gemini-4-argon-high or kimi/k3.
7. Pick the custom model you added in the Chat model dropdown and send a test message.

curl https://api.plan2api.com/v1/chat/completions \
  -H "Authorization: Bearer p2a_your_key" \
  -H "Content-Type: application/json" \
  -d '{"model": "claude/claude-opus-5-5", "messages": [{"role": "user", "content": "Reply with OK"}]}'

Pick your models

The table below lists the model ids you can add to Cursor for Claude Pro and Max subscriptions, ChatGPT plans through Codex, Google AI Pro and Ultra through Antigravity, and Kimi memberships through Kimi Code.

Type the model ids into Cursor exactly as GET /v1/models returns them, including the provider prefix like claude/, openai/, gemini/ or kimi/, because Cursor sends the exact name you enter to POST /v1/chat/completions unchanged. A model only answers if you have connected a subscription for that provider in Plan2API, and one p2a_ key can route across all of your connected subscriptions at once.

What works and what doesn't

• Chat works over the POST /v1/chat/completions endpoint.
• Agent mode can fail on custom endpoints: on June 29, 2026 Cursor staff confirmed that Agent requests can include an ApplyPatch custom tool in the Responses API shape instead of the Chat Completions shape, which OpenAI's API rejects with missing_required_parameter for tools[n].custom, so Agent edits can fail on endpoints that don't accept that tool shape.
• Tab completion never uses your own API key and always keeps using Cursor's own models.
• The Override OpenAI Base URL toggle is global for all OpenAI-family requests rather than per model, so turn it off before switching back to Cursor's built-in OpenAI models (Cursor forum guide, September 13, 2026).
• Cursor routes all own-key requests through its backend servers for final prompt building, which means your custom OpenAI compatible API endpoint must be reachable from the public internet, as https://api.plan2api.com/v1 is.
• Cursor Plus at $8 a month cannot use own API keys at all, Teams and Enterprise plans pay a Cursor Token Rate of $0.25 per 1M tokens on own-key requests to eligible models, and zero data retention does not apply when you use your own keys.

Troubleshooting

Use the table below to match HTTP status codes and error codes from POST /v1/chat/completions to their fixes.

OpenAI models after November 12

OpenAI announced on August 28, 2026 that it will wind down its contract supplying OpenAI models to Cursor, with a proposed shutoff date of November 12, 2026, and noted that Cursor may end access sooner. If you are weighing whether to stay in the editor or move your workflow elsewhere, see our comparison of Cursor vs Claude Code or look at a full Cursor alternative.

OpenAI's help page lists three ways to keep using GPT models in Cursor after the shutoff: putting your own OpenAI API key into Cursor's local Chat and Agent, installing the Codex IDE extension, or routing through an AI gateway such as Amazon Bedrock, Azure or another OpenAI-compatible gateway. With Plan2API, the gateway route lets you use your existing ChatGPT plan in Cursor through live Codex model ids like openai/gpt-6.1-sol, openai/gpt-6-astra and openai/gpt-6-luna without per-token API billing.