---
title: "Connect Cursor to XiuRouter"
description: "Configure Cursor's OpenAI base URL override for XiuRouter, verify Ask mode, and test Agent separately with the selected model."
image: "https://docs.xiu.ai/og.png"
---

> XiuAI 文档索引
> 完整文档索引：https://docs.xiu.ai/llms.txt
> 阅读前先通过索引确认当前可用页面。

# Connect Cursor to XiuRouter

Cursor can send OpenAI-family model requests to XiuRouter through its OpenAI base URL override. The override is global, so record the existing setting and test Ask before relying on Agent.

## Before you start

- Update Cursor to the current stable release.
- Create a dedicated XiuRouter API key for Cursor.
- Select an exact OpenAI-family model ID from the current XiuRouter catalogue.
- Do not add a workstation IP restriction to this key; Cursor can send requests from its own service.

## Configure the connection

1. Open **Cursor Settings → Models**.
2. Record the current OpenAI API key source and base URL when replacing an existing connection.
3. Paste the XiuRouter key into **OpenAI API Key**.
4. Turn on **Override OpenAI Base URL**.
5. Enter:

```text
https://router-api.xiu.ai/v1
```

6. Select **Verify**.
7. Add and select the exact XiuRouter model ID in the same page.

If Cursor already lists the same model ID, select that entry instead of creating a duplicate.

## What the override changes

- Ask requests for eligible OpenAI-family models use `/v1/chat/completions`.
- The override applies globally; Cursor does not provide a different custom provider per model.
- Claude-family and Gemini-family model IDs do not use this OpenAI override.
- Tab completion and features that depend on Cursor-specific models continue through Cursor's own infrastructure.
- Agent depends on Cursor's streamed ApplyPatch tool. A model can answer in Ask and still fail to edit files.

## Verify in stages

1. Ask a short question in **Ask** mode.
2. In XiuRouter's request verification panel, confirm the same key, model, `/v1/chat/completions`, and successful status.
3. Ask **Agent** to inspect one file without changing it.
4. Make one small reversible edit and inspect the diff.

Do not treat the **Verify** button alone as proof that Agent tools work with the selected model.

## Common failures

| Symptom | Action |
| --- | --- |
| Key verification fails | Confirm the key is active, includes the selected model, has the correct group scope, and has available credit |
| Ask works but Agent cannot edit | Update Cursor, create a new conversation, and test a model with more reliable tool calling |
| Cursor's built-in models stop working | Turn off **Override OpenAI Base URL** to return those models to Cursor's routing |
| XiuRouter records no matching request | Confirm the conversation was created after enabling the override and uses an eligible OpenAI-family model |

## Roll back

Turn off **Override OpenAI Base URL**, restore the previous key source and model, and create a new conversation. Confirm traffic has returned to the previous route before revoking the Cursor-specific XiuRouter key.

## Sources and review date

This page was checked against the current XiuRouter Cursor integration and Cursor's public model documentation on August 22, 2026. No real key was used to run a billable Cursor Agent edit during this review.

源文件：https://docs.xiu.ai/en/router/integrations/cursor/index.mdx