Use the Claude API in VS Code

One base URL and one key is the entire integration

For terminal harnesses, use curl -fsSL https://apitoken.sale/setup/connect.sh | bash; Windows PowerShell uses irm https://apitoken.sale/setup/connect.ps1 | iex. The setup writes private files and imports the complete provider model list.

You do not need an Anthropic Console account to run Claude inside VS Code. The free Cline and Continue extensions both accept any Anthropic-compatible endpoint, so the whole job is pointing them at https://router.apitoken.sale, pasting your apiToken.sale key and choosing a model. From that moment Claude answers questions, edits files and reviews diffs without leaving the editor, billed per token from your prepaid balance at a flat 50% off official API pricing.

Nothing exotic happens under the hood. The extension sends ordinary Anthropic Messages requests with an x-api-key header to the gateway; the gateway authenticates your sk-pool-… key, routes the call to the requested Claude model and meters the tokens. Because the wire protocol is untouched, the features these extensions rely on — streaming responses, tool use, large context windows — behave exactly as they would against Anthropic's own endpoint.

See also: Use a Claude API key in Cursor

Set up Cline in four steps

1. 01Install Cline from the VS Code Marketplace and open its settings via the gear icon.
2. 02Set API Provider to Anthropic and enable the custom base URL option.
3. 03Paste https://router.apitoken.sale as the base URL and your sk-pool-••• key below it.
4. 04Enter claude-opus-4-8 as the model, then run a small throwaway task to confirm the key works before trusting it with a real one.

# Cline → Settings
API Provider : Anthropic
Base URL     : https://router.apitoken.sale
API Key      : sk-pool-•••
Model        : claude-opus-4-8

Keep autonomous Cline sessions cheap

Cline is the strong default when you want autonomous edits: give it a task and it reads files, plans, applies changes, runs terminal commands and re-checks its own diff in a loop until the job is done. That loop is exactly why per-token pricing matters — one task can mean dozens of Messages calls, and on a discounted gateway the same session costs half the official rate.

• Prompt caching is billed at the cheaper official cache rates minus your discount, so re-reading the same large files across turns costs a fraction of the naive price.
• Route routine chores to claude-sonnet-5 and reserve claude-opus-4-8 for the problems that genuinely need it — you can switch models mid-session without touching the key.
• Watch the dashboard: it shows token-level usage per request, so a runaway agent loop is visible before it eats the balance.

Point Continue at the same gateway

Continue is the lighter option: it shines at inline chat, quick edits and in-editor help rather than autonomous multi-file runs, and it is configured from a single file you can commit to a repo or sync across machines. Like Cline it is free — only the API usage touches your balance.

# ~/.continue/config.yaml
name: local
version: 1.0.0
schema: v1
models:
  - name: Claude Opus 4.8 (apiToken.sale)
    provider: anthropic
    model: claude-opus-4-8
    apiBase: https://router.apitoken.sale
    apiKey: "{env:APITOKEN_API_KEY}"
    roles: [chat, edit, apply]

Older Continue releases read ~/.continue/config.json instead; the equivalent entry sets "provider": "anthropic", "apiBase": "https://router.apitoken.sale", "apiKey" and "model", and it still works. Keep the autocomplete role on a small dedicated model — Claude earns its tokens on chat, edit and apply.

Cline or Continue: pick by workflow, not by price

Both extensions are free and both bill the same prepaid balance, so the choice is purely about how you like to work. Many developers install both on one key: Cline for delegated tasks, Continue for quick questions and inline rewrites.

The same sk-pool-… key also works simultaneously in Cursor, Roo Code and the Anthropic SDK — one balance across every tool. If you split your time between VS Code and Cursor, the Cursor guide walks the same two-minute flow, and the model catalog lists every model the key unlocks.

Claude API key for Cursor →

Model catalog and per-token prices →

The three errors everyone hits

• 401 Unauthorized: the key or the base URL is wrong. Re-paste the full sk-pool-… key and make sure the base URL is exactly https://router.apitoken.sale, with no extra path segment or typo.
• Model not found: the extension's bundled model list lags behind. Type a current ID explicitly — claude-sonnet-5 or claude-opus-4-8 — instead of picking a stale entry.
• Slow responses or 429: you are over the rate limit. Lower the extension's concurrency and respect the Retry-After header before resubmitting.

If a request fails mid-task in Cline, do not just retry the whole task — the agent keeps its plan, so resuming after fixing the key or model ID avoids paying for the same context twice.

Frequently asked questions

Related guides