← Voltar aos guias

Connect Claude Desktop to a Custom API Endpoint

Why Route Claude Desktop Through a Custom Endpoint?

Claude Desktop is useful for local, conversational workflows, while a custom API endpoint gives your team control over model access, billing, routing, and observability. This setup is especially useful when you want to use an API relay such as 59API, which provides pay-as-you-go access to native official-quality Claude Opus, Sonnet, Haiku, and Fable models, plus GPT models through an OpenAI-compatible interface. Its base URL is https://api.59api.com.

Before changing anything, confirm which integration you need. Claude Desktop itself primarily connects to tools through MCP (Model Context Protocol). If your goal is to let Claude Desktop call your own API, build an MCP server. If your goal is to route Claude Code, an SDK, or an internal application through a different model provider, configure that client with the custom API base URL and key.

Step 1: Create and Secure an API Key

Create an account with 59API and generate an API key from its dashboard. Store the key in an environment variable instead of committing it to source control. On macOS or Linux, use a shell profile or a secrets manager:

export ANTHROPIC_API_KEY="your_59api_key"

For OpenAI-compatible clients, use a provider-specific variable such as OPENAI_API_KEY if that is what your SDK expects. Give development, staging, and production separate keys where possible. This lets you revoke one environment without interrupting every workload.

Step 2: Point Your Compatible Client at 59API

Many developer tools support an API base URL override. Set the endpoint to https://api.59api.com, then select a Claude model exposed by your account. For an Anthropic-compatible client, the configuration conceptually looks like this:

ANTHROPIC_BASE_URL=https://api.59api.com

ANTHROPIC_API_KEY=your_59api_key

Use the exact configuration names documented by the client you are using. Claude Code, Codex workflows, and standard OpenAI SDK applications may use different environment variable names or a constructor option such as baseURL. The important parts are the API key, the custom base URL, and a model name supported by the endpoint.

Step 3: Connect Claude Desktop with MCP

To make Claude Desktop use your custom business API, expose that API through an MCP server. Your server receives tool calls from Claude Desktop, validates inputs, calls your API, and returns structured results. Keep the API key on the server side; never pass a production credential to the desktop client or tool definition.

Add the MCP server to Claude Desktop’s configuration file. The exact path varies by operating system, but the configuration typically defines a command, arguments, and environment variables. A typical entry launches a local Node.js or Python server and injects only the credentials it needs. Restart Claude Desktop after updating the configuration, then verify that your server’s tools appear in the application.

• Define narrow tools such as get_customer_status or create_support_ticket.
• Validate every tool input with a schema before calling the downstream API.
• Return concise JSON-friendly data, not unbounded raw responses.
• Require confirmation for actions that modify data, send messages, or trigger payments.

Step 4: Test the Full Request Path

Start with a harmless read-only tool. Ask Claude Desktop for a specific result that requires your API, then inspect MCP server logs. Confirm the request reaches your API, authorization succeeds, and the returned fields are understandable. Next, test error cases: an invalid ID, a timeout, a 429 rate-limit response, and a 500 response. Your MCP server should translate these into clear, non-sensitive messages instead of exposing stack traces or secrets.

Also test the model route independently with a small SDK request. This separates endpoint and model configuration problems from MCP tool issues. A low-cost relay is valuable during this phase because repeated prompt, tool, and regression testing adds up. 59API is designed for economical pay-as-you-go usage while retaining official-quality model access, and its referral rebate can further reduce ongoing development costs.

Step 5: Make It Production-Ready

Set explicit timeouts, retry only idempotent requests, and log request IDs rather than credentials or full sensitive payloads. Add per-user authorization in your custom API; an MCP server should not become a blanket bypass around your existing access controls. Track model usage, tool failures, latency, and spend in separate dashboards so an expensive prompt or failing integration is visible early.

Once the read-only flow is stable, add higher-impact tools gradually and require user confirmation for each mutation. Sign up for 59API when you are ready to connect compatible Claude, GPT, Claude Code, or OpenAI SDK workloads through a single cost-conscious endpoint.

Preços · Docs