Drop-in for any Anthropic or OpenAI tool.

1. 01
   top up usdc

   Go to /topup, connect a wallet, sign one EIP-3009 transferWithAuthorization. Credits your balance instantly on Base sepolia.
2. 02
   mint a key

   Open /dashboard, hit new key. Shown once — copy it now, we only store an HMAC.
3. 03
   point your tool here

   Set ANTHROPIC_BASE_URL (or OPENAI_BASE_URL) to https://api.x402prompt.com and paste the key. Sections below have copy-pasteable config for every popular tool.

Every request settles in USDC over x402. Pick the mode that fits the caller — same endpoints, same model list, same pricing.

PAYG flow — curl walkthrough

Two requests: a probe to learn the price, then the same payload signed. ThePAYMENT-REQUIRED header carries a base64 JSON quote with the amount, USDC asset address, recipient (payTo), and avalidBefore deadline — sign exactly those fields with EIP-3009.

# 1. probe — no auth, no payment header → gateway quotes a price
curl -i -X POST "https://api.x402prompt.com/v1/chat/completions" \
  -H "Content-Type: application/json" \
  -d '{"model":"claude-haiku-4-5","messages":[{"role":"user","content":"hi"}]}'

# → HTTP/1.1 402 Payment Required
# → PAYMENT-REQUIRED: <base64-json quote: amount, asset, payTo, nonce, validBefore, …>
# → body: {"error":{"code":"payment_required","message":"send PAYMENT-SIGNATURE for …"}}

# 2. sign EIP-3009 transferWithAuthorization with the quoted params,
#    then retry the SAME request with the signature attached:
curl -X POST "https://api.x402prompt.com/v1/chat/completions" \
  -H "Content-Type: application/json" \
  -H "PAYMENT-SIGNATURE: <base64-payload>" \
  -d '{"model":"claude-haiku-4-5","messages":[{"role":"user","content":"hi"}]}'
# → HTTP/1.1 200, plus PAYMENT-RESPONSE header with the settlement tx hash

Reference signer: cmd/test-pay in the repo signs and submits in one shot. Browser clients use wagmi / viem — the dashboard's /topup page is the canonical implementation.

Pricing — flat ceiling per call

Live per-model prices at /models (also at GET /v1/pricing). The quote in PAYMENT-REQUIRED is the same number — the gateway will never charge more than the quoted amount on a given call.

Base URL — replace with your deployment hostname: https://api.x402prompt.com

Auth headers

Two equivalent forms — pick whichever your client defaults to. Both hit the same code path.

Without either header the gateway returns 402 with an x402 PAYG quote. See payment modes for the signature flow.

43+ models across chat, image, and video — Claude, GPT-5.5, Gemini 3.1, Grok, Veo, Kling, nano-banana. Use canonical aliases (they survive provider rotations):

• claude-opus-4-7 — deepest reasoning
• claude-sonnet-4-6 — balanced default
• claude-haiku-4-5 — fast + cheap
• gpt-5.5-pro · gemini-3.1-pro-preview · grok-4.20-beta
• veo-3.1-generate-preview · kling-video-v3 · nano-banana-pro

Full catalog and live pricing at /models. Hit GET /v1/models for the raw JSON.

Two env vars. The official Claude Code CLI honors both — no plugin, no wrapper.

export ANTHROPIC_BASE_URL="https://api.x402prompt.com"
export ANTHROPIC_AUTH_TOKEN="sk-x402-…"
export ANTHROPIC_MODEL="claude-sonnet-4-6"

# then just:
claude "refactor this function to use tanstack query"

If you ran claude /login previously, OAuth credentials override env vars — clear them with claude /logout first.

Codex CLI is OpenAI-shape; point it at /v1/chat/completions via the provider block in config.toml.

# ~/.codex/config.toml
[provider.x402]
base_url = "https://api.x402prompt.com/v1"
api_key  = "sk-x402-…"

[model]
default = "gpt-5.5-pro"
provider = "x402"

Equivalent env form: OPENAI_BASE_URL=https://api.x402prompt.com/v1 + OPENAI_API_KEY=sk-x402-….

Pick Custom endpoint in the provider menu. Base URL ends at /v1 — Hermes appends /chat/completions itself.

# ~/.hermes/config.yaml
provider: custom
base_url: https://api.x402prompt.com/v1
api_key:  sk-x402-…
model:    claude-sonnet-4-6

Works for any model in the catalog, not just Claude — Hermes speaks OpenAI shape across the board.

Simplest path — set the Anthropic env vars and run:

export ANTHROPIC_BASE_URL="https://api.x402prompt.com"
export ANTHROPIC_API_KEY="sk-x402-…"

openclaw

Or wire it through models.json for multi-provider routing:

// ~/.openclaw/models.json
{
  "providers": {
    "x402": {
      "type": "anthropic",
      "base_url": "https://api.x402prompt.com",
      "api_key": "sk-x402-…",
      "models": ["claude-opus-4-7", "claude-sonnet-4-6", "claude-haiku-4-5"]
    }
  },
  "default": "x402/claude-sonnet-4-6"
}

1. Settings → Models → Anthropic API Key — paste sk-x402-…
2. Add ANTHROPIC_BASE_URL=https://api.x402prompt.com to your system environment (macOS: ~/.zshrc; Windows: System Variables).
3. Quit Cursor fully and relaunch — env is read once at startup.

Cursor sends both x-api-key and the Anthropic version header on its own. No further config.

These all expose a Provider dropdown and a Custom Base URL field. Same three values:

Aider CLI: aider --anthropic-api-base https://api.x402prompt.com --model claude-sonnet-4-6 with ANTHROPIC_API_KEY=sk-x402-….

pip install anthropic — same code as direct, just swap the constructor:

from anthropic import Anthropic

client = Anthropic(
    base_url="https://api.x402prompt.com",
    api_key="sk-x402-…",
)

msg = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1024,
    messages=[{"role": "user", "content": "hi"}],
)
print(msg.content[0].text)

npm i @anthropic-ai/sdk — same call shape:

import Anthropic from "@anthropic-ai/sdk";

const client = new Anthropic({
  baseURL: "https://api.x402prompt.com",
  apiKey:  "sk-x402-…",
});

const msg = await client.messages.create({
  model:      "claude-sonnet-4-6",
  max_tokens: 1024,
  messages:   [{ role: "user", content: "hi" }],
});
console.log(msg.content[0].text);

Lets you call Claude / Gemini / Grok / etc. through OpenAI's SDK — useful for existing codebases already wired to openai.

from openai import OpenAI

client = OpenAI(
    base_url="https://api.x402prompt.com/v1",
    api_key="sk-x402-…",
)

# any model in the catalog — not just gpt-*
resp = client.chat.completions.create(
    model="claude-opus-4-7",
    messages=[{"role": "user", "content": "hi"}],
    max_tokens=1024,
)
print(resp.choices[0].message.content)

OpenAI shape:

curl -sX POST https://api.x402prompt.com/v1/chat/completions \
  -H "Authorization: Bearer sk-x402-…" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-sonnet-4-6",
    "messages": [{"role":"user","content":"hi"}],
    "max_tokens": 256
  }'

Anthropic shape:

curl -sX POST https://api.x402prompt.com/v1/messages \
  -H "x-api-key: sk-x402-…" \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-opus-4-7",
    "max_tokens": 1024,
    "messages": [{"role":"user","content":"hi"}]
  }'

Set stream: true. SSE passes through end-to-end — no buffering. Hard wall-clock cap is 240s; client disconnect cancels the upstream request and refunds the reservation delta.

with client.messages.stream(
    model="claude-sonnet-4-6",
    max_tokens=1024,
    messages=[{"role": "user", "content": "tell me a story"}],
) as stream:
    for text in stream.text_stream:
        print(text, end="", flush=True)

curl -NsX POST https://api.x402prompt.com/v1/chat/completions \
  -H "Authorization: Bearer sk-x402-…" \
  -H "Content-Type: application/json" \
  -d '{"model":"claude-sonnet-4-6","stream":true,
       "messages":[{"role":"user","content":"hi"}],"max_tokens":256}'

cache_controlmarkers on Claude requests pass straight through to Anthropic. Cache hits bill at Anthropic's discounted rate; we don't mark up the delta.

{
  "model": "claude-opus-4-7",
  "max_tokens": 1024,
  "system": [
    {
      "type": "text",
      "text": "<long system prompt …>",
      "cache_control": {"type": "ephemeral"}
    }
  ],
  "messages": [{"role": "user", "content": "the actual question"}]
}

At key creation in /dashboard you can set:

• spend limit — hard cap in USDC; once spend + reservation would exceed it, the gateway returns 403 spend_limit_exceeded.
• model allowlist — comma-separated canonical model IDs; anything else gets 403 model_not_allowed.
• revoke — flip the key to revoked and any in-flight request fails fast with 401 key_revoked.

Limits enforce before reservation, so a blocked request never debits balance.

Every response carries X-Request-Id. Include it when reporting issues.