cline2api

Cline's free models as an OpenAI/Anthropic-compatible API — deploy on Cloudflare Workers or run the Python Flask server locally/on any box.

Based on pingmike2/cline2api-workers (reverse-engineered from luawei1/cline2api).

Two runtimes, same API

	Cloudflare Worker	Python (Flask)
File	worker.js	server.py
Runtime	edge, no server	local machine, VPS, Docker, systemd…
Install	npx wrangler deploy	pip install -r requirements.txt
Config	CLINE_REFRESH_TOKEN + API_KEY secrets	same names as env vars

Both expose identical endpoints and identical behavior: dynamic model list, multi-account pool with cooldown rotation, serial request queue, empty-content retry, forced upstream streaming for the free channels.

Free models

Verified working (2026-10-18):

Model	Provider	Status
cline-free/nemotron-3.5-super-vl	NVIDIA (120B)	✅ verified
cline-free/mimo-v2.6-flash	Xiaomi (MoE 309B)	✅ verified
cline-free/muse-spark-1.3-contributor	Meta (multimodal)	✅ verified
cline-free/solar-mini4	35B MoE	✅ verified
deepseek/deepseek-v4-flash	DeepSeek	✅ verified
z-ai/glm-5.3-flash	Z.AI	✅ verified

Known non-free / dead:

• cline-free/deepseek-v4.1-flash (README default of the upstream repo) — retired upstream, returns 404 model not found
• cline-pass/* — requires the paid Cline Pass subscription
• deepseek/deepseek-v4.1-flash (no cline-free/ prefix) — paid tier (402 without credits)

The model list auto-refreshes from Cline every 10 min (merging the official recommended-models free channel), so GET /v1/models is authoritative and picks up new free models automatically.

Deploy to Cloudflare Workers

Option A — button above, then set the secret below in the dashboard.

Option B — CLI:

git clone https://github.com/nullstacks/cline2api
cd cline2api
npx wrangler login
npx wrangler deploy

# set your account token (see next section) and an API key
npx wrangler secret put CLINE_REFRESH_TOKEN
npx wrangler secret put API_KEY

Environment variables (both runtimes)

Variable	Type	Required	Purpose
CLINE_REFRESH_TOKEN	Secret / env	✅	Cline account token(s) — one per line, multi-account pool
API_KEY	Secret / env	recommended	Client access key (default cline2api-default-key; set to empty string to disable auth)

After changing secrets, redeploy (or re-run wrangler deploy) for them to take effect. On the Python side, just restart the process.

Getting a Cline refreshToken (device auth)

Cline's API requires a Cline account token even for free models. Two ways to get one:

Option 1 — Local script (fastest)

python3 cline_oauth.py

1. It prints an authorization URL (https://authkit.cline.bot/device?user_code=XXXX-XXXX)
2. Open it in any browser, log in with Google / GitHub / email, approve
3. The script polls automatically and prints the refreshToken
4. npx wrangler secret put CLINE_REFRESH_TOKEN and paste it — or export it for the Python server (next section)

Option 2 — GitHub Actions (works from your phone)

The repo ships .github/workflows/get-token.yml:

1. In your fork, add repo secrets TG_BOT_TOKEN and TG_CHAT_ID (the flow pushes the login link and the token to your Telegram — the token never appears in Actions logs)
2. Actions → 获取 Cline refreshToken → Run workflow
3. Open the link from Telegram, approve, receive the token
4. Put it in the Worker secret or Python env as above

Notes

• The refresh token is long-lived but can expire — if chat calls start returning 401/403, re-run cline_oauth.py and update the secret/env
• Multi-account pooling: put one token per line in CLINE_REFRESH_TOKEN; when one account's daily free quota runs out (429 Daily free limit reached), the gateway cools it down (parsing Try again in Xh Ym) and switches to the next automatically

Run the Python server

pip install -r requirements.txt          # flask + httpx
export CLINE_REFRESH_TOKEN='<refreshToken from cline_oauth.py>'
export API_KEY='my-secret-key'           # optional

python3 server.py                        # serves on 0.0.0.0:8080 (PORT/HOST to override)

Production options:

gunicorn -w 1 -b 0.0.0.0:8080 server:app          # keep 1 worker: in-memory account pool
flask --app server run --host 0.0.0.0 --port 8080 # dev server

A Dockerfile is included: docker build -t cline2api . && docker run -p 8080:8080 -e CLINE_REFRESH_TOKEN='...' -e API_KEY='...' cline2api.

Usage

curl http://localhost:8080/v1/chat/completions \
  -H "Authorization: Bearer my-secret-key" \
  -H "Content-Type: application/json" \
  -d '{"model":"cline-free/mimo-v2.6-flash","messages":[{"role":"user","content":"hi"}]}'

# Claude Code
export ANTHROPIC_BASE_URL=http://localhost:8080
export ANTHROPIC_API_KEY=my-secret-key
export ANTHROPIC_MODEL=cline-free/nemotron-3.5-super-vl

Browser UA note applies to workers.dev deployments only (Cloudflare edge error 1010) — the Python server has no such restriction.

Behavior notes (both runtimes): max_tokens is stripped (upstream 500s on it for free models); free channels are streamed upstream and aggregated for non-stream clients; requests are serialized (upstream returns empty on concurrency > 1).

Endpoints

GET  /v1/models            model catalog
GET  /v1/health            config diagnostics (account count, auth state)
POST /v1/chat/completions  OpenAI wire (stream + non-stream)
POST /v1/messages          Anthropic wire (auto-converted)

License

MIT — see LICENSE (© luawei1, upstream worker © pingmike2, Python port and packaging © nullstacks)