Using prxy.monster with Cursor

Cursor supports custom OpenAI-compatible endpoints. Set the base URL to prxy.monster and your custom API key to route compatible Chat Completions traffic through prxy.monster.

Configure

1. Open Cursor → Settings → Models
2. Scroll to OpenAI API Key section
3. Click Override OpenAI Base URL
4. Enter:

   https://api.prxy.monster/v1

5. Enter your prxy.monster key in the OpenAI API Key field:

   prxy_live_xxxxxxxxxxxxxxxxxxxxxxxx

6. Click Verify — Cursor will make a test call.

That’s it. Cursor’s compatible custom-model traffic now routes through prxy.monster.

Code change

None. Cursor is an IDE; you don’t touch its code.

Custom model selection

In the Models settings, you can also enable “Custom Models” and add model identifiers that prxy.monster understands. Examples:

• gpt-4o
• gpt-4o-mini
• claude-sonnet-4-6 (prxy.monster translates Anthropic models behind the OpenAI shape)
• gemini-2.0-flash
• bedrock/<model-id> — for AWS Bedrock routing

Verify

curl https://api.prxy.monster/health

Or trigger any Cursor agent action — successful response confirms routing.

What you get

• Pattern memory across all your projects — Cursor’s “agent” mode generates many similar requests across projects. The patterns module learns from successes and re-injects them.
• Semantic cache — repeated “explain this function” / “refactor this” style requests hit cache.
• Cost guards — hard daily cap on Cursor’s API spend (Cursor’s own pricing tier doesn’t enforce this for BYOK users).
• MCP optimization — if you wire MCP into Cursor, the mcp-optimizer module prunes irrelevant tools per request.

Recommended pipeline

PRXY_PIPE=mcp-optimizer,semantic-cache,patterns,ipc

For users on Cursor Pro who want hard cost caps:

PRXY_PIPE=cost-guard,exact-cache,semantic-cache,patterns,ipc

Common issues

• “Cursor uses its own backend by default” — true for the free tier and for default models. The Custom OpenAI Base URL setting only kicks in if you’ve enabled a custom model OR opted to use your own key. Check both.
• “My Cursor Pro subscription stopped counting” — using the Custom OpenAI key path means you BYOK; Cursor stops billing you per-request and your prxy.monster (and underlying provider) bill is the source of truth.
• Inline edits / Composer — both use the same backend wiring as Chat. Same routing applies.

What this does NOT change

Cursor’s local features — fast tab completion, codebase indexing, file search — happen client-side and don’t go through any LLM endpoint. Only the chat / agent / inline-edit calls route through prxy.monster.

Full example

No code example needed — Cursor is a GUI-only setup. The Settings → Models flow above is the entire integration.