Claude Desktop through Model Gate

This is a separate setup from Claude Code. Use a current desktop version that provides third-party inference settings.

Connect

1. Open Help → Troubleshooting → Enable Developer Mode. On Windows, Help is inside the application menu. Restart Claude when prompted.
2. Open Developer → Configure Third-Party Inference.
3. Enter the values below, then apply the configuration using the app's controls.

Do not append /v1 or /v1/messages to the base URL. Use the API key from your account, not a Telegram token or an Anthropic subscription cookie. For a company-specific gateway, use the API address supplied with your account.

When a managed build offers Export, it produces a .mobileconfig for macOS or .reg for Windows. Apply it using your administrator’s deployment procedure. The exported profile may contain your key; do not publish it.

Choose a model and test

Select an available Claude model with Messages, streaming and tool use support. Availability depends on the model and your key's access; this is not a promise that every catalog model works in Desktop. If automatic discovery does not populate the list, use the app's model configuration and the exact model ID shown in your account.

Send one short test message. Inference and connection tests that invoke a model may be billable. The gateway does not turn a Model Gate key into an Anthropic subscription.

When something fails

Check the base URL, key status and selected model's permissions first. For support, send the HTTP status, request ID and a screenshot with all API keys and private content hidden. Do not post a complete configuration export. A connection error is not permission to repeat potentially paid requests blindly.

Official references: Installation and setup, Gateway configuration, Configuration reference. Checked 2026-09-27. These instructions document the supported configuration path; a real Desktop session against your specific route must be tested separately.