Back to guides

Cursor Integration

Point Cursor's plan and agent modes at Kimi K3, GPT-5, Claude, or 200+ models through LLM Gateway. One base URL override, plus what stays on Cursor's backend.

Cursor is an AI-powered code editor built on VS Code. It supports a custom OpenAI base URL, which means you can point its AI panel at LLM Gateway and use any model from our catalog — GPT-5, Claude, Gemini, DeepSeek, or 200+ others — with every request tracked in your dashboard.

One thing up front, because most guides skip it: the base URL override applies to Cursor's AI panel — both plan mode and agent mode. Inline edit (Cmd/Ctrl + K) and tab autocomplete are locked to Cursor's own backend and will not route through any external endpoint. If you want every request in your workflow — including edits and completions — running through LLM Gateway, use Claude Code, Codex CLI, Cline, or OpenCode instead.

Using DevPass? This integration also works with a DevPass plan key. Use root model IDs without a provider prefix (claude-sonnet-4-5, not anthropic/claude-sonnet-4-5) — provider-pinned routing is not available on coding plans; the gateway picks the provider for you.

Quick Start

1. Get your API key

Create an API key in your LLM Gateway dashboard under API Keys.

2. Add the key to Cursor

Open Cursor Settings → Models, then add your LLM Gateway key under OpenAI API Key.

Cursor Settings

3. Override the base URL

In the same Models settings, enable Override OpenAI Base URL and set it to:

1https://api.llmgateway.io/v1

Cursor API Key Input

4. Pick your models

Add any model ID from the models catalog — for example gpt-5, claude-sonnet-4-5, or deepseek-v3.2.

Cursor Model Selection

Open the AI panel (Cmd/Ctrl + L) and every plan and agent request now routes through LLM Gateway.

What works and what doesn't

Cursor feature Routes through LLM Gateway
Chat / plan mode (Cmd/Ctrl + L) ✅ Yes
Agent mode ✅ Yes
Inline edit (Cmd/Ctrl + K) ❌ Cursor backend only
Autocomplete / tab ❌ Cursor backend only

The remaining gaps are a Cursor limitation, not an LLM Gateway one — external OpenAI-compatible endpoints are honored by the AI panel (plan and agent mode), while inline edit and tab completion stay on Cursor's backend.

Model selection tips

  • Provider pinning: prefix the model with a provider to pin it, e.g. openai/gpt-5
  • Discounted models: browse the discounted models and copy the ID
  • Free models: browse the free models
  • Reasoning models: browse reasoning models for planning-heavy work

Troubleshooting

Authentication errors — Verify the API key and that the base URL is exactly https://api.llmgateway.io/v1, and check that your account has credits.

Model not found — Confirm the model ID exists in the catalog and is spelled exactly as shown.

Inline edit or autocomplete still uses Cursor's models — Expected; see the table above.

Need help? Join our Discord.

Why route Cursor through LLM Gateway

  • Any model in plan and agent mode — OpenAI, Anthropic, Google, Meta, DeepSeek, and open-source models through one key
  • Cost tracking — every plan and agent request appears in your dashboard with per-model cost breakdowns
  • Caching — repeated prompts hit the cache instead of the provider
  • One bill — no juggling separate provider accounts

Get started for free — no credit card required.