Configure Claude Desktop with an Inference Gateway

Claude Desktop can connect to a third-party inference gateway through Developer Mode. A Claude key group works directly with an Anthropic-compatible endpoint; an OpenAI-compatible group requires a local route through CC Switch. This workflow runs from choosing the API key group to verifying the request in Dashboard.

Ready to test your chosen route with one short request? Create a BetterToken account

When mapping models in CC Switch, verify GPT, Sonnet, Opus, Fable, and Haiku against the current model ID available to your key.

Connect your own API key through the appropriate route and verify it immediately. BetterToken provides separate OpenAI-compatible and Anthropic-compatible interfaces, so you cannot choose the address from the model name. Open the current BetterToken guide for Claude Desktop, identify your API key group, and follow the corresponding section below.

What third-party inference means in Claude Desktop

Normally, Claude Desktop uses an Anthropic account and Anthropic services. Third-party inference changes the path of the model request: the application sends it to the configured gateway. This does not turn a third-party API into a Claude subscription or remove the application's own requirements for installation, updates, or feature availability.

Do not confuse an inference gateway with Desktop Extensions or MCP. Extensions give Claude access to local applications and data. The gateway routes requests to a model. A failure in one layer does not prove that another layer is broken.

Anthropic's current documentation says Claude Desktop is available for macOS, Windows, and Linux beta. The Help → Troubleshooting and Developer → Configure Third-Party Inference… menu paths below are confirmed by the current BetterToken macOS guide. On another operating system, first check that the installed application version exposes the same settings.

Choose a route based on the API key group

• Claude group: direct Anthropic-compatible gateway; Gateway base URL is https://bettertoken.ai; auth scheme is x-api-key.
• GPT, Kimi, GLM, or another non-Claude provider: use the Claude Desktop tab in CC Switch, OpenAI Responses API (Requires routing), explicit model mapping, and an enabled proxy.

Do not transfer settings from one route to the other. The direct Gateway with x-api-key is for Claude-provider models. For another provider, use the separate CC Switch route: it handles protocol conversion, model mapping, proxying, and the Claude Desktop restart.

Direct setup for a Claude provider

1. Install the current Claude Desktop from the official Claude download page, then enable Developer Mode through Help → Troubleshooting → Enable Developer Mode and quit the application completely.
2. Open Developer → Configure Third-Party Inference…, select Gateway, and enter https://bettertoken.ai as Gateway base URL, your API Key from Token management, and x-api-key as Gateway auth scheme.
3. Click Apply locally, then Save & Restart. After the application opens, select Continue with Gateway; Gateway in the lower-left is the active-route signal.

Advanced route for a non-Claude provider

Use this route for GPT, Kimi, GLM, and other non-Claude providers. Use current Claude Desktop and CC Switch; the current BetterToken guide requires CC Switch v3.16.5 or later.

1. Add a provider on the Claude Desktop tab

In CC Switch, select the Claude Desktop icon in the top bar and click +.

Provider Name: BetterToken-GPT
API Endpoint: https://bettertoken.ai
API Key: your BetterToken API Key

The name only distinguishes providers. Do not publish the key or move it into a shared configuration file.

2. Choose a format and map models

For API Format, select OpenAI Responses API (Requires routing), then click Fetch Models. Map Sonnet, Opus, Fable, and Haiku to the needed current Model ID from BetterToken's model plaza or the Setup window for your key. Enable Declare 1M only when the current catalog says that model has a 1M context window.

3. Apply the provider and restart Claude Desktop

Save the provider, mark it In use, and enable the proxy in the upper-left of CC Switch. Quit Claude Desktop completely and open it again. Gateway in the lower-left confirms the active setup; choose the mapped model beside the input field.

How to verify the request and Dashboard record

Do not begin with workspace-directory access or a long task. Send a short request without personal data—for example, ask for the difference between a JSON array and object—then open the BetterToken Dashboard.

A successful check has two signals:

1. Claude Desktop displays the response and the active Gateway mode.
2. The Dashboard shows a matching record with its time, model, status, input/output/cache tokens, and charge.

If you get a response but no record, confirm that the application is actually using the Gateway rather than another mode. If a record appears with an error status, the route reached the API layer: check the key group, auth scheme, model, and balance. Do not use the Dashboard as storage for a complete prompt or response; remove keys and request contents from logs before contacting support.

What to check when Claude Desktop will not connect

• No Developer menu: enable Developer Mode again and quit Claude Desktop completely instead of merely closing the window.
• Your provider setup needs a fix: check Model list, current IDs, and all required mappings in CC Switch.
• 401 or 403: copy the key again and select x-api-key for a direct Claude route; for the advanced route, verify the key in the CC Switch provider.
• No response and no Dashboard record: for the advanced route, confirm that the intended provider is marked In use and the CC Switch proxy is enabled.
• A record appears but the request fails: check the key group, current model, available balance, and stray spaces in the key.
• Nothing changes after direct setup: click Save & Restart, then select Continue with Gateway.
• Nothing changes after CC Switch setup: quit Claude Desktop completely, reopen it, and check Gateway in the lower-left.

Do not diagnose the Gateway from one interface label. Correlate the Claude Desktop mode, local-route state, and Dashboard request record to see where the chain breaks.

Limitations of this method

BetterToken is an independent API provider, not an Anthropic product or a replacement for a Claude subscription. It is responsible for its own API endpoint and setup documentation. Download access, sign-in, updates, Cowork, Code mode, and other Claude Desktop features depend on the current application version, plan, and Anthropic's rules.

Connecting to the BetterToken API endpoint from Russia does not require a VPN. This statement does not cover the Claude website, application downloads, accounts, or external services. If your menu differs from this article, stop and compare the official Claude Desktop installation documentation with the current BetterToken setup guide.

FAQ

Can I put https://www.bettertoken.ai/v1 in the Claude Desktop Gateway?

No. The direct route for Claude-provider models uses the Anthropic-compatible endpoint https://bettertoken.ai without /v1. For another provider, do not fill the direct Gateway by analogy: configure the provider on CC Switch's separate Claude Desktop tab.

Do I need a Claude account for Gateway mode?

Anthropic's normal instructions ask you to sign in after installation. The current BetterToken Gateway guide takes the user through Continue with Gateway before the normal sign-in. Do not copy steps from one mode into the other; check the interface in your installed Claude Desktop version.

How do I know which auth scheme is correct?

Use x-api-key for a direct Claude route. For another provider, configure the fields in the CC Switch provider according to the current guide. After applying the settings, verify both the Desktop response and a new Dashboard record.

---

Originally published on the BetterToken blog.

BetterToken provides pay-as-you-go access to AI model APIs through
OpenAI-compatible and Anthropic-compatible endpoints — useful if you are wiring
Claude Code, Codex, or your own tooling to a custom base URL.
See the docs to get started.