Run Claude Code at 30% off list price with two env vars

Why Claude Code gets expensive fast

Claude Code is the most productive AI coding tool of 2026, but the token meter is unforgiving. A single multi-hour refactor session on Opus 4.8 can run $5-20 on its own. Multiply that by a full week of work and individual developers easily clear $200/month on tokens. Heavy teams cross $5K/month per 10 developers. The token bill grows linearly with productivity — exactly the dimension you don't want to suppress.

How Anvat fixes it without changing your workflow

Anvat hosts an Anthropic-compatible gateway at api.anvat.app. Point Claude Code at it via ANTHROPIC_BASE_URL and use an Anvat-issued key as ANTHROPIC_AUTH_TOKEN. The wire format is identical to api.anthropic.com — prompt caching, tool use, MCP, vision inputs, streaming, all passthrough. The only thing that changes is the rate: 30% off Anthropic list price, plus 2× credit match on prepaid packs ($50 deposit → $100 of API credit).

Why this beats the obvious alternative

• 30% off metered rates

  Sonnet 4.6 drops from $3/$15 to $2.10/$10.50, Opus from $15/$75 to $10.50/$52.50.

• Full Claude Code feature set

  MCP, hooks, skills, subagents, slash commands, model picker, gateway model discovery — all work unchanged.

• Prompt caching passthrough

  Stacks with the discount. Claude Code's 60-85% typical cache hit rate is preserved.

• Two env vars to switch

  ANTHROPIC_BASE_URL + ANTHROPIC_AUTH_TOKEN. Reversible in 60 seconds.

• $2 free on signup

  No card required. Enough to run a few hundred small Sonnet requests as a smoke test.

Quickstart

Drop these two lines into ~/.zshrc or ~/.bashrc

# Anvat — discounted Anthropic-compatible gateway
export ANTHROPIC_BASE_URL=https://api.anvat.app/v1
export ANTHROPIC_AUTH_TOKEN=sk-anvat-...

# Optional: let Claude Code populate /model picker from gateway models
export CLAUDE_CODE_ENABLE_GATEWAY_MODEL_DISCOVERY=1

# Verify
claude "say 'gateway' and stop"

FAQ

Does prompt caching work through Anvat?

Yes. Anvat is a passthrough gateway — your cache_control blocks are forwarded to Anthropic unchanged. You get the full 5-minute cache TTL and the same 10%-of-input cached read rate. Discount applies on top of the cached price.

Will my MCP servers still work?

Yes. Anvat doesn't touch MCP — it operates at the LLM API layer. Your MCP servers run locally inside Claude Code as before, and tool_reference blocks are forwarded correctly to the gateway.

What happens if Anvat goes down?

Failover is one env-var change. Unset ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN, set ANTHROPIC_API_KEY to your direct Anthropic key, and Claude Code falls back to api.anthropic.com. We publish a status page and monitor 24/7 — 99.997% uptime over the last 90 days.

Is this allowed by Anthropic's TOS?

Yes. Anthropic explicitly supports ANTHROPIC_BASE_URL for gateway routing — it's documented in the official Claude Code docs under LLM Gateway Configuration. Anvat pays Anthropic directly for the underlying capacity and operates as a billing/discount layer on top.

Can I use both my direct Anthropic key and Anvat?

Yes — they're separate keys, set independently. Most teams keep both for failover: ANTHROPIC_BASE_URL + ANTHROPIC_AUTH_TOKEN for normal traffic, unset both to fall back to ANTHROPIC_API_KEY direct.