> ## Documentation Index > Fetch the complete documentation index at: https://docs.hostlinku.com/llms.txt > Use this file to discover all available pages before exploring further. # Claude Desktop > Connect Claude Desktop to the Hostlinku gateway using Configure Third-Party Inference, without an additional local proxy. [Claude Desktop](https://claude.com/download) can use Hostlinku through **Configure Third-Party Inference** in app versions that expose this menu. Enter your base URL and API key in the app, then select a model available for your key's Service Mode. This guide covers the Desktop app. For the terminal or VS Code/JetBrains extensions, see [Claude Code](https://docs.hostlinku.com/en/integrations/claude-code.md). This inference connection does not require an additional local proxy or changes to MCP configuration. ## Before you start * Install Claude Desktop from the [official download page](https://claude.com/download). * Create a Hostlinku API key in [Console → API Keys](https://console.hostlinku.com/api-keys/) with a Service Mode that provides your chosen Claude model. * Ensure your USD balance and key spending limit are sufficient. Usage is billed by Hostlinku; see [Billing & Balance](https://docs.hostlinku.com/en/billing.md). What has been tested The Windows screenshots below show successful model discovery and an inference test. A simple conversation in Claude Desktop's **Code** mode also succeeded with **Fable 5.2**. This does not verify every Chat, Cowork, artifacts, connector, or model feature. Menu names and locations can vary by app version. ### Enable Developer Mode On Windows, open **☰ → Help → Troubleshooting → Enable Developer Mode**. If the Developer menu does not appear after enabling it, fully quit and reopen the app. ![Claude Desktop on Windows: Enable Developer Mode in Help → Troubleshooting](https://mintcdn.com/hostlinku/D_n6COs9jkVA6WFt/images/integrations/claude-desktop-developer-mode.webp?fit=max\&auto=format\&n=D_n6COs9jkVA6WFt\&q=85\&s=8d52f77f1d224acf4278f03098e0a0ee) ### Open third-party inference settings Open **☰ → Developer → Configure Third-Party Inference…**. ![Claude Desktop on Windows: Configure Third-Party Inference in the Developer menu](https://mintcdn.com/hostlinku/D_n6COs9jkVA6WFt/images/integrations/claude-desktop-third-party-menu.webp?fit=max\&auto=format\&n=D_n6COs9jkVA6WFt\&q=85\&s=2b29c13ba052454fe69fabbb47bc78ac) ### Configure the Hostlinku connection In **Connection**, use these settings: | Field | Value | | - | - | | Provider | `Gateway` | | Credential kind | `Static API key` | | Gateway base URL | `https://api.hostlinku.com` | | Gateway API key | Your Hostlinku API key, for example `sk-your-api-key` | | Gateway auth scheme | `x-api-key` | Enter the raw key without a `Bearer ` prefix. Do not append `/v1` or `/messages` to the base URL. Hostlinku already provides `GET /v1/models` for discovery and `POST /v1/messages` for Anthropic Messages inference. The pictured test uses `x-api-key`. Hostlinku also accepts `Authorization: Bearer`, but the screenshots do not test that option. Custom inference headers are unnecessary for this basic configuration. ![Claude Desktop gateway configuration: Hostlinku base URL, masked static key, x-api-key, and successful Test connection results](https://mintcdn.com/hostlinku/D_n6COs9jkVA6WFt/images/integrations/claude-desktop-gateway-settings.webp?fit=max\&auto=format\&n=D_n6COs9jkVA6WFt\&q=85\&s=13d5e81a7e0ccf512c2f8141d09dd096) ### Test the connection and discover models Click **Test connection**. A successful result displays **Model discovery** and **Inference**. To check only the list, use **Test model discovery** in **Models**, if available. The screenshot's **12 models** is the result for that key at the time of testing, not a fixed total for all accounts. The list follows the key's Service Mode and model availability. If you enter a **Model list** manually, use slugs from `GET /v1/models` that support Anthropic Messages; do not copy another provider's model list. ### Save and try a conversation Click **Save Changes**, then **Apply Changes** if available. Follow any restart prompt. Open a **Code** session in Desktop, choose an available model, and send a simple message such as `hello`. The screenshot below shows a response with **Fable 5.2**. Also check **Console → Request Logs** to correlate the request with its API key, model, status, and cost. Test connection itself can consume tokens and incur a small usage charge. ![Claude Desktop Code mode: a successful greeting with Fable 5.2 in the HostlinkuGateway environment](https://mintcdn.com/hostlinku/D_n6COs9jkVA6WFt/images/integrations/claude-desktop-code-conversation.webp?fit=max\&auto=format\&n=D_n6COs9jkVA6WFt\&q=85\&s=4dca2b60c563d61ae63c22850eb8eddb) ## Feature boundaries * **Inference is separate from MCP.** This setting routes model requests to Hostlinku; connector or MCP settings do not automatically change the inference provider. * **Gateway-managed Claude Code** and **Server-side Auto mode checks** are not required for the static-key setup shown here. This integration has not tested those features. * Tool calling, vision, reasoning, context, and caching depend on the model and protocol path. See [Anthropic Messages](https://docs.hostlinku.com/en/anthropic-messages.md) and the [model catalog](https://docs.hostlinku.com/en/models.md). * Successful discovery does not guarantee every listed model supports all Desktop agent features. Start with a Claude model and a simple task. ## Troubleshooting | Symptom | Check | | - | - | | Developer or third-party inference menu is missing | Confirm that your Desktop version provides the feature, enable Developer Mode, and fully restart the app. | | `401` | Check the key, revocation/expiration status, and auth scheme. Do not prepend `Bearer ` in the key field. | | Empty discovery or model not found | Run Test model discovery; check `GET /v1/models` with the same key and a slug available in its Service Mode. | | `400 MODEL_NOT_IN_SERVICE_MODE` | Use a key whose Service Mode provides that model. | | `402` | Check the available USD balance and key spending limit. | | `429` | Follow `Retry-After` when provided, reduce concurrency, and see [Usage & Limits](https://docs.hostlinku.com/en/usage-and-limits.md). | | Test connection succeeds but a Code session fails | Check the selected model and request details in Console; a short connection test does not verify full tool/agent support. | Do not share your API key or configuration exports containing credentials. When requesting help, include the Desktop version, model, Service Mode, and request ID without the key. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.