Claude API · Private Access

Claude models,
on your
own terms

OpenAI-compatible endpoint. Drop-in for Claude Code, Cursor, Cline, or any SDK. Your dedicated key. Your private usage dashboard. No shared quota. No surprises.

# Two lines. You're in.
export ANTHROPIC_BASE_URL="https://api.mayfly.wiki/v1"
export ANTHROPIC_API_KEY="your-key"

# Claude Code
claude "refactor this service"

# Or direct API
curl https://api.mayfly.wiki/v1/messages \
  -H "x-api-key: your-key" \
  -d '{"model":"claude-sonnet-4-6","messages":[...]}'
5
Claude models
800/wk
Requests included
0
Data logged
2vars
To integrate
The Problem

Why shared proxies
break at the worst
moment

01
Shared quota, shared pain
One heavy user exhausts the pool. Everyone gets 429s. Your users blame you — not the proxy you're routing through.
02
Black-box rate limits
No Retry-After header, no usage endpoint. You can't tell if it's the proxy or Anthropic upstream that's throttling you.
03
Model bait-and-switch
Request Opus 4, get a cheaper model dressed up as Opus. Unverifiable identity, no audit trail, no recourse.
04
Opaque data handling
Your prompts traverse unknown infrastructure. No data residency commitment, no way to verify who has access.
05
Single-credential fragility
One upstream account gets banned and the whole service goes dark. No failover, no notice, no SLA — just silence.
06
Zero budget visibility
Monthly surprise invoices. No per-key breakdown, no weekly burn rate. Finance can't plan, and you can't optimize.
The Mayfly Approach

One key.
All the control.

Isolated quota per key
Your limits are yours alone. No other customer's usage touches your weekly or monthly quota. Keys revoked instantly on request.
Transparent rate limiting
Every response carries X-RateLimit-Remaining-Weekly headers. 429s include exact Retry-After seconds and per-window usage so you can handle backpressure correctly.
Private usage dashboard
Access your personal stats at /usage/<first-32-chars-of-key> — no password needed. Weekly and monthly burn, per-model breakdown, quota bars.
Zero data logging
Requests pass through and are not stored. Rate limiting runs on request counts only, not content. Your prompts are never written to disk.
Multi-credential failover
Multiple upstream Anthropic credentials in rotation. If one hits its quota ceiling, the next takes over automatically — no service interruption, no customer impact.
OpenAI-compatible wire format
Change two environment variables. Same request shape, same response shape. Tested and working with Claude Code, Cursor, Cline, Python SDK, and raw curl.
Available Models

Latest Claude.
Always.

All models live on the same endpoint. Switch by changing the model field — no config change, no re-auth, no latency difference.

Model ID
Tag
Best for
Context
claude-fable-5
Newest
Maximum capability, complex reasoning, frontier agentic tasks
200K
claude-opus-4-8
Powerful
Deep reasoning, code review, long-context document analysis
200K
claude-sonnet-4-6
Recommended
Balanced speed and quality — daily coding, writing, structured analysis
200K
claude-haiku-4-5-20251001
Fast
Highest throughput — autocomplete, classification, summarization at scale
200K
claude-3-7-sonnet-20250219
Legacy
Backward compatibility for workflows pinned to a specific 3.x version
200K
Quickstart

Up in two
minutes.

Step 01
Get your key
Email [email protected]. You'll receive a personal API key within 24 hours along with your usage dashboard URL.
Step 02
Set two env vars
Point your client at the endpoint. No SDK changes, no code changes — just the base URL and your key. Works with every client that supports a custom base URL.
Step 03
Pick your model
Any model ID in the list works. Start with claude-sonnet-4-6 for the best balance of speed and quality.
# Claude Code
ANTHROPIC_BASE_URL=https://api.mayfly.wiki/v1 \
ANTHROPIC_API_KEY=your-key \
claude "fix this bug"
# Python SDK
import anthropic
client = anthropic.Anthropic(
  base_url="https://api.mayfly.wiki/v1",
  api_key="your-key"
)
client.messages.create(model="claude-sonnet-4-6", ...)
# curl
curl https://api.mayfly.wiki/v1/messages \
  -H "x-api-key: your-key" \
  -H "content-type: application/json" \
  -d '{"model":"claude-sonnet-4-6","max_tokens":1024,"messages":[{"role":"user","content":"Hello"}]}'
Pricing

Simple.
Transparent.

Pay once per month. Your key expires automatically — renew to continue.
No per-token billing, no hidden fees, no meter anxiety.

Startup Program
Freeforever
if your idea is a fit
  • Unlimited access while building
  • All 5 Claude models
  • Higher rate limits
  • Direct feedback channel
  • No strings attached
800
per week
resets Monday 20:00 CET/CEST
3,200
per month
resets 1st of each month
heavy requests
>30K input tokens counts double

Limit exceeded → HTTP 429 with Retry-After header and exact window usage in the body. Never a silent failure.

Startup Program

Building something
interesting?
Tell us.

We're looking for early-stage startups and indie builders with real ideas. Send us a short note about what you're working on — if it's a fit, you get free API access while you build. No pitch deck required. No committee. Just email.

01
Send your idea
Email [email protected] with what you're building and how you'd use Claude. A few sentences is enough.
02
We evaluate quickly
We look for ideas we find genuinely interesting or useful — AI tooling, research, developer tools, creative projects. If it's doable and interesting, you're in.
03
Free access, no strings
You get a dedicated key with full access to all models. Build what you want. We only ask that you tell us how it goes.
Apply

Email [email protected] with the subject line "Startup Access" and a brief description of your idea. If it's something we can support, we'll send a key the same day.

Get Started
Ready when you are.

Email us and you'll have a working key within 24 hours. Two env vars and you're calling Claude — no onboarding call, no dashboard signup, no drama.

[email protected]  →