📝

Cursor

AI-first IDE pointed at clawfeeder via OpenAI-compatible endpoint

Protocol: OpenAILast verified 2026-05-19

Cursor is a VS Code fork built around inline AI. Its 'Override OpenAI Base URL' setting lets you redirect every chat / cmd-K / autocomplete call to clawfeeder.ai while keeping the OpenAI-style request shape.

Prerequisites

Cursor 0.40 or later, and a clawfeeder API key (cf-sk-***).

Steps

1Open Cursor settings

Press Cmd/Ctrl + Shift + J to open Settings, or navigate Cursor → Settings → Cursor Settings → Models.

2Override OpenAI Base URL

In the OpenAI API key section, toggle 'Override OpenAI Base URL' and paste the clawfeeder endpoint. Then paste your cf-sk-*** in the OpenAI API Key field.

Base URL: https://clawfeeder.ai/v1
OpenAI API Key: cf-sk-***your_key***

3Add custom models

Cursor's default model list won't know about clawfeeder's catalog. In Models → Custom Models, add the names you want — e.g. gpt-5.2, claude-opus-4-7, claude-sonnet-4-6. They appear in the model picker after a restart.

Custom Models:
  gpt-5.2
  gpt-5.5
  claude-opus-4-7
  claude-sonnet-4-6
  claude-haiku-4-5-20251001

4Verify in chat

Open the Cursor chat panel (Cmd/Ctrl + L), pick one of your custom models, and send 'hi'. A reply means the gateway is wired up.

Verify

Pre-flight check from terminal — ensures clawfeeder side accepts your key before debugging Cursor settings.

curl -X POST "https://clawfeeder.ai/v1/chat/completions" \
  -H "Authorization: Bearer cf-sk-***" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-5.2","messages":[{"role":"user","content":"say READY"}],"max_tokens":10}'

Expected response includes: "content":"READY"

FAQ

Cursor Tab completion still uses Cursor's own model — does that work?

No. The lightweight tab/autocomplete model is hosted by Cursor and not exposed via Base URL override. Only chat / cmd-K / agent calls hit clawfeeder.

Anthropic models in Cursor — do they work?

Yes via clawfeeder's OpenAI-compatible endpoint. Add claude-* model names under Custom Models. Cursor wraps them as OpenAI calls; clawfeeder translates internally to Anthropic format upstream.

Why does Cursor say 'Model not found'?

Two causes: (1) the model name doesn't match clawfeeder's catalog — check /docs#models for exact IDs. (2) The custom model wasn't saved or Cursor needs a restart after adding it.

Can I use Cursor agent mode with clawfeeder?

Yes — agent mode uses the same Base URL setting. It will burn credits faster than chat because of multi-turn planning; watch /dashboard during heavy sessions.

Don't have an API key yet?

Get started free →