Run Claude in Cursor without an Anthropic account

Yes — Cursor only needs a base URL and a key

You do not need an Anthropic account to run Claude in Cursor. Cursor's Anthropic provider lets you override the base URL and paste your own API key, and apiToken.sale issues a key that Cursor accepts in exactly that slot. Sign up, paste two values into Settings, and Claude answers inside Cursor with no Anthropic involvement at any point.

The reason this works is that Cursor talks to the Anthropic Messages API: a POST to /v1/messages with an x-api-key header and an anthropic-version header. apiToken.sale exposes exactly that API at its router, so Cursor cannot tell the difference — it sends the same request shape it would send to Anthropic and gets the same response shape back. Streaming, tool use and system prompts all behave the standard Anthropic way, because the protocol on the wire is the standard Anthropic protocol.

Requests travel straight from your machine to the endpoint you configured. There is no relay through Cursor's servers for BYOK traffic and no extra moving part to debug: if the endpoint answers, Cursor works.

See also: Use a Claude API key in Cursor

Point Cursor's Anthropic provider at apiToken.sale

1. 01Open Cursor → Settings → Models and scroll to the Anthropic API section.
2. 02Set the base URL to https://router.apitoken.sale and paste your apiToken.sale key (it looks like sk-pool-•••) into the API key field.
3. 03Add a current model ID to the model list — claude-opus-4-8 is the safe default — and make sure the toggle next to it is on.
4. 04Open a chat, select that model, and send any message. A streamed reply confirms the key, the base URL and billing are all live.

# Cursor → Settings → Models → Anthropic API
Base URL : https://router.apitoken.sale
API key  : sk-pool-•••
Model    : claude-opus-4-8

# Optional: verify the endpoint before you even open Cursor
curl https://router.apitoken.sale/v1/messages \
  -H "x-api-key: sk-pool-•••" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{"model":"claude-opus-4-8","max_tokens":64,"messages":[{"role":"user","content":"ping"}]}'

The curl check is worth the thirty seconds it costs. If it returns a JSON completion, every later problem is inside Cursor's settings; if it returns an auth error, the key itself is wrong and no amount of clicking in Cursor will fix it.

What runs on your key inside Cursor — and what does not

Bringing your own Anthropic key reroutes the features that call a Claude model. It does not change how Cursor's own features are delivered, and it does not unlock anything Cursor gates behind its own plan.

• Chat, Composer and agent mode run on the Claude model you selected, and the tokens bill against your prepaid balance.
• Inline edit (Cmd/Ctrl+K) uses the same selected model and the same key.
• Cursor Tab autocomplete is served by Cursor's own autocomplete models, not the Anthropic API — your key is never involved, and Tab availability still depends on your Cursor plan.
• Features Cursor reserves for its own subscribers stay reserved; a model provider key changes where model calls go, not what your Cursor license includes.

The common confusion: Claude answers in chat but Tab suggestions are gone. That is expected — Tab never used your Anthropic key, even when the key came from Anthropic itself. The two systems have separate billing and separate providers.

One key, the whole Claude lineup

A single apiToken.sale key unlocks the full Claude line — Opus, Sonnet and Haiku — so you can switch tiers inside Cursor without juggling credentials. Add each model ID in Settings → Models and pick per task:

Because all three draw from the same balance, the practical workflow is to default to Sonnet, drop to Haiku for throwaway prompts, and reserve Opus for tasks where a wrong answer costs more than the tokens do.

Current model lineup with per-model pricing →

Prepaid billing instead of an Anthropic invoice

There is no Anthropic account on the other side, so there is no Anthropic invoice either. You top up an apiToken.sale balance by card or crypto, and every request from Cursor decrements it per token at a flat 50% off official API rates. Instant access: the key works as soon as it is generated, with no waitlist and no usage-tier review.

• Token-level usage per key in the dashboard, so you can see exactly what Cursor costs you per day.
• An optional lifetime spending limit and expiration date per key — issue a dedicated key for Cursor, cap it, and a runaway agent loop can never burn more than you allowed.
• The same balance also covers GPT, Gemini and Kimi models if you later point other tools at it; nothing is locked to Cursor or to Claude.

Estimate a month of Cursor usage in the free calculator →

From signup to the first answer

1. 01Create an apiToken.sale account — sign up with Google or GitHub to start with $5 of platform bonus credit (email/password accounts do not receive the bonus).
2. 02Top up by card or crypto when you need more balance; the bonus credit is enough to validate the whole setup first.
3. 03Generate an API key (sk-pool-…) and, if you want a hard ceiling, set its lifetime spending limit and expiration date.
4. 04Paste the base URL and key into Cursor as shown above, select claude-opus-4-8, and send your first prompt.

Nothing about how you use Cursor changes after this. You simply source the key and the balance from apiToken.sale instead of Anthropic — and you never create an Anthropic account at all.

The three setup errors people actually hit

• 401 Unauthorized: the key was pasted truncated or with a stray space, or you edited the OpenAI provider instead of the Anthropic one. Re-paste the full key in the Anthropic section.
• Model not found: the model ID is not in Cursor's model list or is outdated. Add the exact string claude-opus-4-8 and enable it.
• Verify button fails: the base URL is wrong. It must be the bare router origin — no /v1 suffix, no trailing path — because Cursor appends the Messages API path itself.

If chat works but responses stop mid-stream on very long agent runs, check the key's lifetime spending limit in the dashboard first — an exhausted or expired key fails exactly this way.

Frequently asked questions

Related guides