Complete Guide to Connecting Third-Party APIs in Claude Code Desktop

Introduction

Claude Code Desktop is Anthropic's local AI programming tool, but the official API pricing can be prohibitively expensive for many developers. This guide walks you through configuring third-party APIs in Claude Code Desktop, enabling flexible switching between Claude and GPT models while significantly reducing costs.

Downloading and Installing Claude Code Desktop

Claude Code was initially released as a command-line tool (CLI), offering code generation, debugging, and refactoring capabilities for terminal users. In mid-2025, Anthropic launched the desktop version, wrapping the CLI's powerful features in a graphical interface. The official API uses a per-token billing model — taking Claude Opus 4 as an example, input tokens cost $15 per million and output tokens cost $75 per million. For individual developers who use it frequently, monthly expenses can easily reach hundreds of dollars. Third-party API platforms typically reduce costs by 50%-80% through bulk purchasing and regional pricing differences.

Start by visiting the Claude Code official website to download the installer. Once downloaded, open it to begin installation — the entire process takes about two and a half minutes and is straightforward. Just follow the prompts step by step.

After installation, click "Start" and the system will prompt you to log in. Instead of following the official login flow, we'll use Developer Mode to configure a third-party API.

Enabling Developer Mode and Configuring Claude API

Enabling Developer Mode

Developer Mode is a built-in advanced configuration entry point in Claude Code Desktop, similar to Chrome's Developer Tools. Once enabled, the application exposes its underlying API routing configuration, allowing users to customize the target endpoint for requests. This design is common in Electron-based desktop applications — the app itself acts as an HTTP client making requests, and Developer Mode allows you to modify the Base URL, redirecting traffic to any compatible API server.

Here are the steps to enable it:

Click the menu in the upper-left corner
Navigate to Help → Troubleshooting → Enable Developer Mode
When the confirmation dialog appears, select Enable
Wait for Claude Code to restart

Configuring the Third-Party API

After restarting, click the upper-left menu and navigate to Developer → Configured Third-Party Interface to access the configuration page:

Credential Kind: Select Static API Key
API Key: Enter the key obtained from your third-party platform
Base URL: Enter the endpoint address provided by the third-party platform
Auth Scheme: Select X-API-Key

Regarding the authentication scheme selection, it's important to understand the difference between the two common types: Bearer Token is an authentication method defined in the OAuth 2.0 standard, placing the token in the Authorization field of the HTTP request header; X-API-Key is Anthropic's custom authentication header, placing the key in the x-api-key request header. Since we're connecting to a third-party service compatible with the Claude protocol, we select the X-API-Key scheme.

Adding a Model

Click "Add Model" and enter the Claude model name you want to use in the Model ID field (e.g., claude-opus-4-8). Note that this method only supports Claude series models — both the model ID and display name must be filled in correctly.

Once complete, click "Apply" to save and restart. Send a test message like "Hello" — if you get a normal response, the third-party API configuration for Claude models is successful.

Connecting GPT Models via a Proxy Tool

Calling Claude models can be expensive. If you want to use GPT models (such as GPT-4 or GPT-5.5) within Claude Code, you'll need a request proxy tool to handle protocol conversion.

Why a Proxy Tool Is Needed

Claude Code Desktop natively only supports requests in Claude's protocol format. To connect GPT models, you need a middleware layer that converts Claude-format requests into OpenAI format — that's what the proxy tool does.

From a technical perspective, while both Claude API and OpenAI API are RESTful interfaces, they differ significantly in request body structure. Claude uses the /v1/messages endpoint with a system field and specific content block format; OpenAI uses the /v1/chat/completions endpoint with a messages array and role field structure. The proxy tool is essentially a locally-running reverse proxy server that listens on a local port, receives Claude-format requests, parses and reassembles them into OpenAI format, forwards them to the target API, then converts the response back to Claude format before returning it to the Claude Code client. This pattern is known as a Protocol Adapter in API gateway and microservices architecture.

Proxy Tool Configuration Steps

Download the proxy tool: Download Claude API Proxy from GitHub (the specific URL can be found in relevant community forums)
Extract and run: Double-click to run cloud-api-proxy.hs — a command-line window will appear
Open the configuration page: Hold Ctrl and click the second link displayed in the command-line window
Add a provider: Click "Provider" → "One-Click Add" and enter your third-party platform's GPT API Key
Set up model mapping: In "Default Model Mapping," select the GPT model you want to use (e.g., GPT-4, GPT-5.5)

Updating the Configuration in Claude Code

Go back to the proxy tool's "Connection Info" page:

Copy the Base URL provided by the proxy tool and replace the Base URL in Claude Code's configuration
Copy the API Key provided by the proxy tool and replace the key in Claude Code
Click "Apply Changes" and restart

Send a test message — if you get a normal response, the GPT model connection is successful.

Important Notes

Don't close the proxy tool: When using GPT models, the command-line window must remain running. Closing it will prevent API calls from working
Response display quirks: Due to Claude Code's built-in prompt mechanism, responses may show the model as Claude even though GPT is actually being called. You can verify the actual model used through your third-party platform's usage logs
Switching models: To switch between Claude and GPT, you'll need to update the Base URL and API Key configuration

Comparison and Recommendations

Configuration Method	Supported Models	Proxy Tool Required	Complexity
Direct third-party API	Claude series	No	Low
Via proxy tool	GPT series	Yes	Medium

For daily use, choose the appropriate model based on your task type:

Code generation and refactoring: Claude models excel here
General conversation and analysis: GPT models offer better value
Long-context tasks: Choose based on each model's context window

Regarding context window selection strategy: the context window refers to the maximum number of tokens a model can process in a single conversation. As of 2025, Claude Opus 4 supports a 200K token context (approximately 150,000 words), GPT-4 Turbo supports 128K tokens, and GPT-4o also supports 128K. For large codebase refactoring tasks, a larger context window means the model can understand dependencies across more files simultaneously. However, longer contexts mean higher API costs, so in practice you'll need to balance comprehension depth against budget.

Conclusion

By leveraging Developer Mode and third-party API configuration, you can flexibly use different AI models in Claude Code Desktop — preserving Claude Code's excellent programming interface while significantly reducing costs. Although the setup involves multiple steps, following this guide should get you up and running on the first try. When using third-party API services, make sure to choose a stable and reliable platform, and always keep data security in mind.