Cursor

Use Fusion Model with Cursor's custom OpenAI provider settings.

Compatibility notice: Cursor officially limits custom API keys to standard chat models. Its OpenAI Base URL override is global, and Cursor can send request shapes that differ from standard Chat Completions. Basic chat may work, but Agent tool calls, streamed tool arguments, reasoning output, and future Cursor updates may cause errors or incomplete behavior.

Fusion configuration

What is Cursor?

Cursor↗ is an AI code editor with chat, code editing, and agentic workflows.

Cursor provides an Override OpenAI Base URL setting for custom OpenAI-compatible endpoints. Use this setting to connect Cursor to Fusion.

Setup

Step 1: Get your FUMO API key

Follow Getting started for online self-service accounts (see Enterprise onboarding for enterprise accounts) to create and store your API key. If you use FUMO and have a sales representative, enter their Sales Code at signup or bind it later as described in Console and account.

Step 2: Configure the OpenAI provider

1. Open Cursor Settings.
2. Go to Models, then expand API Keys.
3. Enable OpenAI API Key and paste your FUMO API key.
4. Enable Override OpenAI Base URL.
5. Enter:

https://capi.fumolab.ai/api/fusion/v1

The Base URL is a prefix. Cursor should append /chat/completions when it sends a standard OpenAI chat request.

The override applies globally rather than per custom model. Disable it before switching back to Cursor-managed models. Requests using your API key still pass through Cursor's backend for final prompt construction before reaching the configured endpoint.

Step 3: Add Fusion Code

In Cursor's model settings, add:

fusion-code-1.0

Select the model in Chat or Agent and begin with a simple text prompt before testing edits or terminal tools.

You can also add the authorized Fusion Max model ID:

fusion-max-1.0

Compatibility limitations

• Agent tools: Cursor may flatten or otherwise transform tool definitions and tool calls. Fusion expects the standard Chat Completions structure.
• Streaming tools: partial tool-call arguments may not be reconstructed correctly without a Cursor-specific adapter.
• Model verification: Cursor may validate against assumptions made for built-in OpenAI models or model-list endpoints that Fusion does not currently expose.
• Reasoning metadata: provider-specific reasoning fields may be ignored or rejected.
• Managed models: while the OpenAI Base URL override is enabled, Cursor may send built-in OpenAI model requests to the override as well. Disable the override when returning to Cursor-managed models.
• Future client updates: Cursor can change its private request format independently of the Chat Completions standard.

If ordinary chat succeeds but tool use fails, check the tool-call format and client compatibility first. Use OpenCode for the more direct Chat Completions integration.

Troubleshooting

Authentication error

Confirm that you are using an API key created in the console and that you copied it in full. Check the model ID. For a 403 response, follow the access checks in Troubleshoot the first call.

404 from the API

Set the override to exactly https://capi.fumolab.ai/api/fusion/v1. Do not add /chat/completions unless the Cursor version you are using explicitly requests a full endpoint.

Cursor rejects the model ID

Confirm that the exact authorized ID is present in Cursor's model settings. Cursor may reject it before sending the request, and the public gateway may reject an unknown or unauthorized ID before it reaches the Fusion core.

Chat works but Agent fails

This usually indicates a request-shape or tool-call compatibility issue. Cursor integrations may require a dedicated normalization layer; Fusion does not currently provide one.

Resources

• Cursor website↗
• Cursor API key documentation↗
• Fusion Chat Completions API

Configure Claude Code through CC Switch

Supported models: fusion-max-2.0, fusion-max-1.0, fusion-code-1.0. Only Max 2.0 supports low / medium / high / max effort (default medium); the other two models do not support adjustment.