How do I configure a third-party API in Cursor?

Steps — open Settings (Cmd/Ctrl + Shift + J) → Models on the left → turn off the built-in default models → Add Model and enter a custom model ID (e.g. `claude-opus-5`, `gpt-5.6-sol`, `glm-5.2`) → turn on OpenAI API Key and paste your `sk-` key → turn on Override OpenAI Base URL and enter `https://api.lmuai.com/v1` → click Verify → press Cmd/Ctrl + L to open the Chat panel and switch to that model. Note that **only the Chat / Plan panel uses the third-party API**; Cursor's native features like Composer / Tab autocomplete still use the official backend.

Which third-party API protocol does Cursor support?

Only the **OpenAI-compatible protocol** (`/v1/chat/completions`), overridden via Settings → Models → Override OpenAI Base URL. There is currently no native "Override Anthropic Base URL" toggle; to use Claude models in the Chat panel, add the Claude model ID as a custom model over the OpenAI-compatible endpoint, and the LMU AI backend performs the OpenAI ↔ Anthropic protocol translation.

Is a third-party relay cheaper for Cursor?

Usually yes. A relay bills by actual token usage, which is more cost-effective than a Cursor Pro monthly subscription for heavy users (especially frequent Opus / GPT-5 usage); one key also serves Claude / GPT / GLM / Qwen / DeepSeek at once. Note that Cursor's Composer / Tab autocomplete cannot use a third-party API and still consume the Cursor subscription quota.

Can Cursor use a third-party relay API?

Yes, but **only for the Chat / Plan panel**. Cursor's Settings → Models has an "Override OpenAI Base URL" toggle; once on, requests from the Chat / Plan panel (Cmd/Ctrl + L) go through the third-party relay, while native agent features like Composer / Tab autocomplete / Apply / Inline Edit stay locked to the Cursor official backend.

How do I use Claude / GLM / Qwen in Cursor?

Open Settings → Models (Cmd/Ctrl + Shift + J) → add a custom model (model ID `claude-opus-5` / `claude-sonnet-5` / `glm-5.2`, etc.) → turn on Override OpenAI Base URL → set Base URL to `https://api.lmuai.com/v1` → set the API Key to the `sk-` key generated in the LMU AI console.

Can Cursor's Composer / Tab autocomplete use LMU AI?

No. Cursor's core agent features — Composer, Tab autocomplete, Apply, Inline Edit — are locked to the Cursor official backend and cannot be taken over by a third-party relay. For a "full-capability" LMU AI experience, install the [official Codex extension](/en/docs/tools/vscode-plugin) or the [Claude Code for VS Code extension](/en/docs/tools/claude-code-vscode) in Cursor (Cursor is compatible with the VS Code extension marketplace).

Verify fails when I add a model?

Common — non-official OpenAI model names (e.g. `claude-opus-5`, `glm-5.2`) do not pass Cursor's connection check. Workaround: first add an official OpenAI name like `gpt-4` to pass Verify (confirming the Base URL + key connect), then add the custom model IDs you want separately; Verify warns but **it runs fine in practice**.

Do I need a VPN to use LMU AI with Cursor?

The part of the Chat panel configured for LMU AI does not — `api.lmuai.com` is a direct domestic connection. But Cursor's account sign-in, Composer, and Tab autocomplete use the Cursor official backend, which may still need network optimization in China.

Configure Cursor with a third-party API

Configure a third-party API in Cursor — turn on Override OpenAI Base URL, enter your sk- key, and use Claude, GPT, and more in the Chat panel.

Cursor can point the Chat / Plan panel's requests at a third-party relay API via Override OpenAI Base URL. This page explains how to use that native method to connect to LMU AI.

Video walkthrough

Why there is no Base URL field under Cursor's Anthropic setting — a ~6 min video on Bilibili (in Chinese) that walks through the entire setup on this page.

Key limitation: only the Chat / Plan panel is affected

Cursor's Override OpenAI Base URL only affects the Chat / Plan panel (the conversation opened with Cmd/Ctrl + L). Cursor's core agent features — Composer, Tab autocomplete, Apply, Inline Edit — are all locked to the Cursor official backend and cannot be taken over by a third-party relay.

If you need a "full-capability" agent experience in Cursor (including Composer-type features), install instead (Cursor is compatible with the VS Code extension marketplace):

Official Codex extension — a Codex agent over the LMU AI OpenAI protocol
Claude Code for VS Code extension — a Claude Code agent over the LMU AI Anthropic protocol

Configuration steps

Step 1 — Open Cursor Settings

Press Cmd + Shift + J (macOS) or Ctrl + Shift + J (Windows / Linux) to open Settings, and click Models in the left sidebar.

Step 2 — Turn off the default models (recommended)

Cursor ships with many default models (GPT-4 / Claude-3.5, etc.); leaving them enabled conflicts with custom routing. In the model list on the Models page, turn them all off (toggle grayed out on the left).

Step 3 — Add custom models

Click Add Model and enter the LMU AI model IDs you want, for example:

claude-opus-5
claude-sonnet-5
gpt-5.6-sol
glm-5.2
qwen3.8-max-preview
deepseek-v4-pro
kimi-k3

Add each model once. For the full list of available models, see the Model Gallery.

Step 4 — Turn on Override OpenAI Base URL

Scroll to the OpenAI API Key area on the Models page:

Turn on the OpenAI API Key toggle
Paste the LMU AI console key (starts with sk-) into the API Key box
Turn on the Override OpenAI Base URL toggle
Set Base URL to:
```
https://api.lmuai.com/v1
```
Click Verify to check the connection

A failed Verify doesn't mean it's misconfigured

Cursor's Verify checks using the model name you entered, and non-official OpenAI names (claude-opus-5, glm-5.2, etc.) won't pass. Workaround: add a gpt-4 first to pass Verify (confirming the Base URL + key connect), then add the custom model IDs you want separately; Verify warns but it runs fine in practice.

Step 5 — Switch model and use it

Press Cmd/Ctrl + L to open the Chat panel → pick the custom model you added in Step 3 from the model dropdown at the top → start chatting, and requests will go through LMU AI.

Common model IDs

Model ID	Description
`claude-opus-5`	Claude Opus 5 (flagship, recommended)
`claude-sonnet-5`	Claude Sonnet 5 (balanced)
`claude-haiku-4-5`	Claude Haiku 4.5 (high-speed)
`gpt-5.6-sol`	GPT-5.6 Sol
`glm-5.2`	Zhipu GLM-5.2
`qwen3.8-max-preview`	Qwen 3.8 Max Preview
`deepseek-v4-pro`	DeepSeek V4 Pro
`kimi-k3`	Kimi K3
`claude-opus-4-8`	Claude Opus 4.8 (previous generation)
`claude-sonnet-4-6`	Claude Sonnet 4.6 (previous generation)
`gpt-5.5`	GPT-5.5 (previous generation)
`glm-5.1`	Zhipu GLM-5.1 (previous generation)

For the full list, see the Model Gallery.

Common issues

Want Composer / Tab autocomplete to use LMU AI too?

Native Override OpenAI Base URL can't do it — these Cursor core features are locked to the official backend. To route everything through LMU AI, use:

Official Codex extension — an OpenAI-protocol agent
Claude Code for VS Code extension — an Anthropic-protocol agent

Cursor is compatible with the VS Code extension marketplace, so both extensions install directly in Cursor.

Does the Base URL need `/v1`?

Yes. Enter https://api.lmuai.com/v1. Cursor appends /chat/completions for a final request to https://api.lmuai.com/v1/chat/completions. Omitting /v1 gives a 404.

Built-in Claude 3.5 errors out after turning on Override?

This is a known Cursor routing conflict — turning on the OpenAI Base URL override interrupts the Anthropic branch. To keep using Claude in the Chat panel, add the Claude model ID (e.g. claude-opus-5) as a custom model, and the LMU AI backend does the OpenAI ↔ Anthropic protocol translation.

401 / 403 errors?

Check in order:

Is the API Key the sk- key generated in the LMU AI console (not an official OpenAI key)?
Is the Base URL https://api.lmuai.com/v1 (must include /v1)?
Is the model ID one LMU AI supports (see the Model Gallery)?
Does your current plan / billing group include the model you're calling?

Does BYOK affect my Cursor subscription?

BYOK usage does not consume Cursor's monthly fast-request quota
Core features like Composer / Tab autocomplete still consume the Cursor subscription quota (because they don't use BYOK)
To drop the Cursor subscription entirely, see the Codex / Claude Code extension approach above

Last updated: 2026-09-10