Paste one line for free Claude — or self-host a rotating pool. 39 free doors, ready to paste.
Skip the “free Claude key” hunt — that’s the dead-end. The base_url is the real lock. Point Claude Code / Cline / Roo / Cursor at a free door and it just runs. Below is the whole kit — grab the line you need. 
━━━━━━━━━━━━━━━━━━━━━━━
Grab-and-paste — real Claude, no card
Drop one line into Claude Code (with that door’s own free key):
ANTHROPIC_BASE_URL=https://api.poe.com/v1 # Poe — 300 free Claude pts/day, no card
ANTHROPIC_BASE_URL=https://aihubmix.com # AiHubMix — standing free test quota (native)
ANTHROPIC_BASE_URL=https://ai-gateway.vercel.sh # Vercel — $5 free every 30 days
ANTHROPIC_BASE_URL=https://api.z.ai/api/anthropic # z.ai — ~$3/mo + trial, GLM-4.6 ≈ Claude-tier
CLAUDE_CODE_USE_VERTEX=1 # Google Vertex — $300 / 90 days (real Claude)
CLAUDE_CODE_USE_BEDROCK=1 # Amazon Bedrock — ~$200 new-account (real Claude)
No terminal? In Cline / Roo or SillyTavern pick “OpenAI Compatible,” paste any base_url + key — that box IS the trick.
That’s the starter kit. The full trove 
— 
rare · 
no-terminal · 
native Anthropic (no glue)
🎯 REAL Claude doors — free / trial credits
├─ Poe — 300 pts/day on real Claude, no card · api.poe.com/v1
├─ z.ai GLM plan — ~$3/mo + trial, native api.z.ai/api/anthropic (GLM-4.6 ≈ Claude-tier coding)
├─ Vercel AI Gateway — $5 free / 30 days, zero markup · ai-gateway.vercel.sh
├─ Amazon Bedrock — ~$200 new-account credits · CLAUDE_CODE_USE_BEDROCK=1
├─ Google Vertex — $300 / 90 days · CLAUDE_CODE_USE_VERTEX=1
├─ copilot-api — a GitHub Copilot seat (free for students/OSS) → Claude endpoint · npx copilot-api start --claude-code
├─ claude-proxy — your own ChatGPT / Gemini sub drives Claude Code at $0/call
└─ CLIProxyAPI — OAuth-pool your own Claude/Codex/Gemini CLI seats into one auto-rotating Anthropic endpoint
🆓 Free open-model doors — Claude-tier (add the glue below for Claude-only tools)
├─ Puter.js — keyless, no signup: one <script> + puter.ai.chat(...)
├─ OpenRouter — one signup → free key; append :free to ids · openrouter.ai/api/v1
├─ GitHub Models — free, no card, any GitHub PAT · models.github.ai/inference
├─ Hugging Face Router — free HF token, ~1k req/day · router.huggingface.co/v1
├─ NVIDIA NIM — 1000–5000 never-expire credits, no card, 100+ models · integrate.api.nvidia.com/v1
├─ Chutes.ai — TAO-subsidized, some models genuinely $0, no card · llm.chutes.ai/v1
├─ ModelScope 魔搭 — Alibaba, ~2000 free calls/day · api-inference.modelscope.cn/v1
├─ g4f.dev hosted — nothing-to-run base_urls (g4f.space/api/groq …), key secret
├─ 
gpt4free — keyless local server (g4f api → :1337); uptime = coin-flip, fallback only
└─ DuckDuckGo→OpenAI — self-host; carries real claude-3-haiku over DDG’s free chat
🧩 The glue — one repo makes ANY door speak Anthropic /v1/messages
├─ claude-code-router — the 36k
standard; ccr code runs Claude Code on any backend, per-task fallback
├─ y-router — zero-deploy: ANTHROPIC_BASE_URL=https://cc.yovy.app + OpenRouter key (shared worker archived — deploy your own to depend on it)
├─ claude-code-proxy — clean Python shim: any OpenAI base_url → Anthropic /v1/messages
├─ anthropic-proxy-rs — fast Rust shim, full streaming + tool-calls + thinking blocks
├─ claude-adapter — interactive wizard, no config-editing, wraps any OpenAI provider
├─ OmniRoute — one install stacks 40+ free-forever tiers + a Claude Code fallback profile
└─ cerebras gateway — free Cerebras key = 1M tokens/day, no card, ~2000 tok/s in Claude Code
🔁 Never-dry kit — pool free keys, auto-rotate, even auto-farm credits
├─ LiteLLM — reference proxy; native /v1/messages, load-balance + fallbacks + 429 cooldown
├─ 9router — 40+ providers, 3-tier subscription→cheap→free fallback + token compression
├─ gpt-load — key-pool proxy that keeps native Anthropic passthrough
├─ uni-api — smartest failover: re-ranks keys by their real 72h success rate
├─ Bifrost — Go, native /anthropic drop-in, weighted rotation, ~50× faster than LiteLLM
├─ freellmapi — stacks 29 providers / ~4B tokens/mo behind one endpoint
├─ all-api-hub — browser extension that auto-farms daily free credits + health-checks keys
├─ metapi — folds all your relay accounts into one Claude-Messages key
├─ new-api · CRS · Sub2API — self-host CN relay panels + 拼车 seat-share (split one Claude Max seat, with billing/quota)
└─ AiHubMix — standing free testing quota on real Claude · native aihubmix.com
Two honest notes: free-tier amounts churn (Poe pts · Vercel $5 · Bedrock ~$200 · Vertex $300 · NVIDIA “never-expire”) — confirm at signup; and gpt4free/g4f uptime is a coin-flip (fallback, not foundation). Its all free tiers, trial credits, your own subs, and open endpoints.
The key was never the lock. Own the base_url and Claude is just a setting.
!