Cline API Setup Guide (2026): Connect Claude, GPT & Gemini in VS Code for Less

To use Cline with a custom API, install the Cline extension in VS Code, open its settings, choose OpenAI Compatible as the API Provider, then enter a Base URL, an API key, and a model ID such as anthropic/claude-sonnet-4-6. This lets Cline call Claude, GPT, or Gemini through any OpenAI-compatible gateway instead of billing directly through Anthropic or OpenAI.

Why Cline Users Look for a Custom API

Cline (the VS Code extension formerly known as Claude Dev) reads your project files, proposes diffs, runs terminal commands, and iterates autonomously on multi-step coding tasks. It ships with support for several providers out of the box, but most developers eventually hit the same three friction points:

• Billing sprawl. Running Cline against Anthropic directly means a separate account, a separate invoice, and (for many users outside the US) a card that has to pass Anthropic's billing checks.
• Token cost. Agentic tools like Cline re-send large chunks of file context on almost every turn. Without prompt caching, that context gets billed at full input-token price on every single request.
• Model lock-in. Testing whether Claude Sonnet or GPT-5.5 handles a refactor better shouldn't require re-authenticating with a different provider.

A custom OpenAI-compatible endpoint solves all three: one key, one bill, and free choice of model per request.

Step-by-Step: Configuring Cline

1. Open VS Code, go to Extensions (Cmd+Shift+X / Ctrl+Shift+X), search Cline, and install it.
2. Click the Cline icon in the sidebar, then open the gear/settings icon inside the Cline panel.
3. Under API Provider, select OpenAI Compatible.
4. Fill in the three required fields:

5. Save, then send a test prompt in the Cline chat panel. If you get a normal response, the connection works.

If your gateway also exposes a native Anthropic-compatible path, some Cline versions let you pick an "Anthropic Provider" mode instead of OpenAI Compatible - this preserves full tool-use behavior and is worth trying if you rely heavily on Claude's function calling inside Cline's agentic loop.

Cutting Cost: Prompt Caching Matters More in Cline Than Anywhere Else

Because Cline resends system prompts, file trees, and open-file context on nearly every step of an agent run, prompt caching has an outsized effect on your bill. Anthropic's prompt caching can cut repeated input-token cost by roughly 70-90% once a cache hit occurs, since cached tokens are billed at a fraction of the normal input rate instead of full price.

Two things to check when picking a gateway for Cline:

• Does it pass through prompt caching? Some relays strip cache-control headers or don't forward them upstream, silently costing you the discount.
• Is pricing per-model or blended? A gateway that marks up every model equally will erase any savings you'd get from routing cheap tasks to a cheaper model.

Common Errors and Fixes

A Cheaper, Card-Free Way to Run Cline

If you're outside the US and don't want to fight Anthropic's or OpenAI's card verification just to try Cline, Safa API is worth pointing your Base URL at. It's a single OpenAI-compatible endpoint that routes to Claude, GPT, and Gemini with official, non-throttled model access - so Cline behaves exactly as it would against the native APIs, just cheaper and easier to pay for.

What makes it a good fit specifically for Cline's usage pattern:

• Lower per-token pricing than calling Anthropic or OpenAI directly, which adds up fast given how many calls an agentic loop makes.
• Prompt caching supported, so Cline's repeated file-context sends actually get the discount instead of being billed at full price every turn.
• No credit card required - sign up and top up without US billing verification.
• Alipay accepted, so you can pay in RMB without a forex card.
• One endpoint, one key for Claude, GPT, and Gemini - switch the Model ID field in Cline instead of re-configuring providers.

Setup takes the same three fields as above: point Base URL at Safa API, drop in your key, pick a model ID, and Cline is ready to go.

常见问题 / FAQ

Does Cline support any OpenAI-compatible API, or only official providers?

Cline's "OpenAI Compatible" provider mode works with any endpoint that implements the standard chat completions API, including third-party gateways. You just need a Base URL, an API key, and a valid model ID.

Will using a third-party API break Cline's tool-use / file-editing features?

Not if the gateway correctly forwards tool-call requests and streaming responses. Function calling is what lets Cline apply diffs and run commands, so test a multi-step task (not just a single chat message) after switching providers.

Is prompt caching automatic, or do I need to configure it in Cline?

Prompt caching is typically handled by the model provider and gateway based on request structure, not by a Cline setting. The main thing to verify is that your chosen gateway actually passes caching support through rather than discarding it.