🧑‍💻

Cline

VS Code coding agent (formerly Claude Dev) wired to clawfeeder

Protocol: BothLast verified 2026-05-19

Cline is an open-source VS Code extension for autonomous coding tasks. It supports multiple providers; the 'OpenAI Compatible' provider lets it call clawfeeder directly.

Prerequisites

VS Code with the Cline extension installed (search 'Cline' in marketplace), and a clawfeeder API key.

Steps

1Open Cline settings

Click the Cline icon in the activity bar, then the gear ⚙ in the top-right of the Cline panel.

2Choose OpenAI Compatible provider

Under API Provider, select 'OpenAI Compatible'. Fill in Base URL and API Key fields.

API Provider:    OpenAI Compatible
Base URL:        https://clawfeeder.ai/v1
API Key:         cf-sk-***your_key***
Model ID:        claude-sonnet-4-6

3Configure model + context

Set Model ID to any clawfeeder-supported model. For autonomous coding tasks, claude-sonnet-4-6 or claude-opus-4-7 give the best results. Set Context Window to match the model (200K for claude-* family, 1M for gpt-5.x).

4Save and test

Click Done. Type a task in the Cline input — e.g. 'list the files in this directory and propose a refactor'. Cline will call clawfeeder for each thinking/action step.

Verify

Smoke test from terminal with the same provider format Cline uses.

curl -X POST "https://clawfeeder.ai/v1/chat/completions" \
  -H "Authorization: Bearer cf-sk-***" \
  -H "Content-Type: application/json" \
  -d '{"model":"claude-sonnet-4-6","messages":[{"role":"user","content":"reply READY"}],"max_tokens":20}'

Expected response includes: "content":"READY"

FAQ

Should I use the Anthropic provider or OpenAI Compatible?

OpenAI Compatible. The Anthropic provider expects sk-ant-*** keys and Anthropic-only model IDs. clawfeeder accepts cf-sk-*** through both protocols, but Cline's OpenAI Compatible flow is more flexible (any model in the catalog works, not just Anthropic ones).

Cline shows 'Request failed' — how to debug?

Open VS Code Developer Tools (Help → Toggle Developer Tools), check the Console for the actual HTTP error. Most commonly: 401 (wrong key), 400 (model ID typo), or 503 (clawfeeder upstream unavailable).

Why does Cline burn credits so fast?

Each Cline task involves many tool-use round trips (read file → think → write → re-read → think again…). Long tasks easily hit 50+ requests. Use claude-haiku-4-5-20251001 for cheaper iteration loops, switch to opus only for hard refactors.

Does Cline's vision/image input work?

Yes for vision-capable models (claude-opus-4-7, claude-sonnet-4-6, gpt-5.4). Cline encodes images as base64 in the OpenAI message format; clawfeeder passes them through unchanged.

Don't have an API key yet?

Get started free →