Kilo Code Setup

Connect QCode.cc in the Kilo Code VS Code extension: Settings -> Providers -> Custom provider, with Provider API set to Anthropic Messages for Claude

Last verified: 2026-09-18 · 📄 Per official docs (Kilo Code extension; documentation at kilo.ai)

⚠️ Ownership and UI changed: on 2026-07-15 Anaconda announced its acquisition of Kilo Code. Since then Kilo reworked the custom-endpoint entry: it is no longer Roo Code's "API Provider dropdown + tick Use custom base URL", but Settings → the Providers tab → the Custom provider dialog (stated verbatim on the official anthropic and openai-compatible pages). The steps below follow the new UI.

At a glance¶

Kilo Code is an open-source AI coding extension for VS Code. It is a fork of Roo Code, which in turn forked from Cline — same ancestry, diverged interfaces: Kilo uses the Custom provider dialog in the Providers tab, while Roo Code and Cline still use an API Provider dropdown. Do not copy these steps into those two tools; read their own pages.

Which protocol¶

🔴 Never call Claude through OpenAI Compatible — QCode's OpenAI endpoint rejects Claude models with model_not_available_on_endpoint. The official docs likewise put Anthropic and MiniMax under Anthropic Messages.

Install¶

Search for Kilo Code in the VS Code marketplace, or follow the instructions on kilo.ai.

Configure Claude (Custom provider)¶

1. Open Kilo Code's sidebar Settings (the gear icon) and go to the Providers tab
2. Scroll to the bottom and click Custom provider
3. Fill in the dialog:

1. Click Submit; the models appear in the model picker
2. Send a message in the chat to verify

Mainland China: swap the host for https://asia.qcode.cc/api (Asia node, near Korea / Taiwan / Hong Kong); the key stays the same. The official Base URL examples are written in versioned form (something like https://api.your-provider.com/v1), and Kilo auto-fetches the model list from a valid URL. For the Anthropic Messages leg we suggest stopping at /api and letting the extension append /v1/messages. No trailing slash. Whether an extra slash actually produces a 404 we have not tested, so follow the suggested form. Whatever you set in the UI is stored in a config file (the official pages call it kilo.json in one place and kilo.jsonc in another) and can be edited directly.

Configure GPT and Chinese models (OpenAI Compatible)¶

The same Custom provider dialog, only two fields change:

1. Set Provider API to OpenAI Compatible
2. Base URL: https://api.qcode.cc/openai/v1
3. API key: the same cr_ key
4. Models: type gpt-6-sol, gpt-5.6-terra, or a Chinese model id such as glm-5.2

For the Responses-style interface of GPT / xAI the official value is OpenAI Responses; Anthropic models still only work through Anthropic Messages (Base URL https://api.qcode.cc/api). The two legs are not interchangeable.

Verify connectivity¶

Check the endpoint and key with curl first, so you know whether to debug the extension's UI:

KEY="cr_your_qcode_key"
curl -X POST https://api.qcode.cc/api/v1/messages \
  -H "x-api-key: $KEY" -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{"model":"claude-sonnet-5","max_tokens":32,"messages":[{"role":"user","content":"ping"}]}'

JSON containing content means the endpoint side is fine; if the extension still fails, the problem is in its settings.

Troubleshooting¶

Related¶

• Endpoints & API Paths — protocol × model-family table
• Roo Code Setup — upstream project, nearly identical setup
• Cline Integration — further upstream
• Chinese Models