Same CLI, same workflow: set two environment variables and Claude Code runs through NeuralBox. Pay per use instead of $20–200/month, plus 300+ models to switch between.
Claude Code is the best AI coding tool in the terminal — but official access has failure modes: accounts get banned for "suspicious activity" or regional flags, and Pro/Max plan limits run out at the worst possible moment, locking you out until the next reset.
NeuralBox provides a drop-in Anthropic-compatible API proxy: point two environment variables at our endpoint and Claude Code keeps working exactly as before — same CLI, same workflow, no account with Anthropic required. Pay by card (Stripe) or crypto, strictly per use — you buy tokens that never expire instead of paying $20–200 a month for a capped subscription.
A bonus Anthropic doesn't offer: the /model command switches Claude Code to any of 300+ models — GPT, Gemini, DeepSeek, Qwen, Grok. Cheap models for routine chores, a flagship for the gnarly refactor.
Grab your sk-nb-… key in your profile after signing up — then pick any of these.
| Endpoints | /api/v1/messages (Anthropic format), /api/v1/chat/completions (OpenAI format) |
| Authentication | Bearer sk-nb-… or x-api-key |
| Claude models | current Sonnet/Opus, transparent model-name mapping |
| Other models | 300+ via /model or the model parameter |
| Streaming | SSE, long reasoning sessions supported (up to 20-minute timeout) |
| Status line | balance/model/cost — installed with one command |
| Compatibility | Claude Code, Anthropic SDK, Roo Code, Cline and others |
| Scenario | On NeuralBox | Official Anthropic |
|---|---|---|
| Typical Claude Code request (Sonnet) | ≈ $0.05–0.30 | Pro subscription $20/mo with limits |
| Routine chores via /model (Qwen, DeepSeek) | under a cent | not available |
| Intensive coding day | ≈ $3–10 | Max $100–200/mo |
| Occasional use (a couple of times a week) | ≈ $2–5/mo | still $20/mo |
Pay per use: tokens are charged per request, the price depends on the model and context size. Token rate shown for the Basic plan ($5 = 150,000 tokens); larger plans cut token cost by up to 40%. Anthropic subscriptions for comparison: Pro $20/mo, Max $100–200/mo.
Requests are served by real Claude models through official aggregator API channels. NeuralBox is not affiliated with Anthropic — we provide compatible proxy access.
Nothing: same CLI, same commands, same format. Only two environment variables change. Hooks, MCP servers and project settings keep working.
Yes, that's one of the main use cases: the proxy doesn't depend on your Anthropic account. Sign up on NeuralBox, set the key — and you're back to work in a couple of minutes.
There are no "messages per 5 hours" windows here: you're limited only by your token balance. Heavy sessions just cost what they cost, with no forced cooldowns.
Depends on intensity: a Sonnet request costs cents, an active coding day roughly $3–10. With occasional use it's far cheaper than $20/mo; with daily heavy use it's comparable to Max — but with no limits and no account risk.
300+: GPT (OpenAI), Gemini (Google), DeepSeek, Qwen, Llama, Mistral, Grok and more. Switch with /model right inside a session — handy for running routine tasks on cheap models.
Yes, SSE streaming is fully supported, including long reasoning sessions of flagship models (up to a 20-minute timeout per response).
Yes: the same key works with the Anthropic SDK and with OpenAI-compatible tools (Roo Code, Cline) via /v1/chat/completions.
Install the statusline with one command — your balance and session cost show right inside Claude Code. Full per-model, per-day breakdowns are in your dashboard.
Sign up, grab a key, set two variables — and you're coding again. No subscription, no account at risk.
Get a key