Skip to main content
This guide explains how to integrate StepFun models with Cursor for code generation, file editing, and other development tasks in Chat and Agent.

​
Requirements

​
Cursor Account Requirements

An active paid Cursor subscription (such as Pro or Teams) is required for BYOK (Bring Your Own Key). Free / Hobby accounts cannot use the method in this guide to call Step models through Cursor’s built-in Chat or Agent.

​
Install Cursor

Download and install the editor from the Cursor website. Cursor 3.15.20 or later is recommended.

​
Choose a Billing Method

  • Standard API (pay-as-you-go): Confirm that your account has available balance or credits and permission to use the model.
  • Step Plan (subscription quota): Confirm that your Step Plan subscription is active, has available quota, and provides access to the model.

​
Obtain an API Key

Create an API key on the StepFun platform API keys page.

​
Configure Cursor

This configuration applies to Chat and Agent. Code completion (Tab) continues to use Cursor’s built-in models.
Override OpenAI Base URL is a global override. When enabled, requests using the OpenAI protocol are sent to the custom endpoint. Before using models included in your Cursor subscription, disable the override and start a new conversation.
1

Open Model Settings

Open Cursor Settings and go to Models. You can also search for Cursor Settings in the command palette.
2

Configure the Connection

Enter the following connection settings on the Models page. The table uses the Step Plan base URL as an example:Set OpenAI API Key to your Step API key for request authentication.
Select the base URL for your billing method:

Standard API (pay-as-you-go): https://api.stepfun.ai/v1
Step Plan (subscription quota): https://api.stepfun.ai/step_plan/v1

Do not include /chat/completions in the base URL; Cursor appends this path automatically.
3

Add a Custom Model

Add a custom model on the Models page and set Model ID to step-5-preview, step-3.7-flash, step-3.5-flash-2603, or step-3.5-flash.step-5-preview has a 1M-token model context window.
4

Select the Model and Start a New Conversation

Open Chat or Agent, select the Step model you added, and start a new conversation to test the integration.

​
Verify the Integration

​
Text Connectivity

In a new conversation, enter:
Expected result: The model returns OK without a 401, 403, or protocol error.

​
Agent Tool Calls

Use a separate test directory on a machine with Python installed. Create a README.md file containing Cursor Step Plan test, then open that directory in Cursor. In a new Agent conversation with the Step model selected, enter:
Review the proposed file operations and commands before granting any required permissions. Expected results:
  1. The conversation shows actual tool calls to read README.md, create hello.py, and execute a command.
  2. hello.py exists in the working directory, and you can inspect its contents and changes in the editor.
  3. The script outputs Hello, world!.
A code snippet returned in chat alone does not verify Agent tool execution.

​
Request Routing

In the Step platform usage or request records, confirm that the model ID matches your selection in Cursor and that the billing channel matches the configured standard API or Step Plan base URL.

​
Restart Verification (Optional)

Quit Cursor completely and reopen it. Confirm that the Models settings are retained, then repeat the text connectivity test in a new conversation with the same model.

​
Switch Back to Cursor Models

Changing the model selector alone does not disable the custom base URL.
  1. In Cursor Settings → Models, disable Override OpenAI Base URL.
  2. Disable Use OpenAI API Key if needed.
  3. Start a new conversation and select the Cursor model you want to use.

​
Frequently Asked Questions

First, check the subscription for the account currently signed in to Cursor in the Cursor Dashboard:
  • Free / Hobby: Free accounts do not support BYOK. Meet the Cursor account requirements first.
  • Active paid subscription: Confirm that you are signed in to the subscribed Cursor account and have selected the custom Step model in a new conversation. For team accounts, also confirm that the administrator allows BYOK and check team usage limits. Check your Step account’s model permissions and balance or subscription quota for the selected billing method.
Check the following in order:
  1. Override OpenAI Base URL is enabled and the base URL matches your billing method: https://api.stepfun.ai/v1 for the standard API or https://api.stepfun.ai/step_plan/v1 for Step Plan.
  2. The base URL does not include a full API path such as /chat/completions.
  3. The API key is valid and Use OpenAI API Key is enabled.
  4. The model ID is spelled correctly and your account has access to the model through the selected channel.
  5. Your network can reach https://api.stepfun.ai. If you use a proxy, check its configuration as well.
Check that:
  • The key was copied completely, contains no extra spaces or line breaks, and has not been revoked.
  • The key belongs to your international (.ai) account, not another site or provider.
  • Use OpenAI API Key is ON.
Paste the complete key, enable the toggle, and retry in a new conversation. For permission or quota errors, check your account against the requirements for the selected billing method.
  1. Confirm that the model ID is correct, such as step-5-preview, and that your account has access to it through the selected channel.
  2. Confirm that the override is enabled and the base URL matches the selected billing method.
  3. Select that custom model in a new Chat or Agent conversation.
  • Confirm that you are using Agent mode, not only a text conversation mode.
  • Select the custom Step model rather than Auto, then retry the minimal Agent task in a new conversation.
  • Check whether Cursor is waiting for permission to edit files or execute terminal commands. The Python example also requires a working local Python interpreter.
  • Distinguish model request failures from tool execution failures. A missing Python installation or denied command does not indicate a failed model connection.
If the issue persists, provide the Cursor version, model ID, time of the failure, redacted error details, and Cursor Request ID when contacting support. To copy the Cursor Request ID, open the … menu in the top-right corner of the affected Chat or Agent conversation and select Copy Request ID.