XThe x402 API for AI Agents
Data your software can buy itself. No signup. No API key. No account.
Every endpoint answers with HTTP 402 and a machine-readable price. Your agent pays in stablecoin and retries — one round trip, no human in the loop. Settlement is verified on-chain before a single byte is returned.
Endpoints
| /api/price/{mint} | $0.001 |
| /api/compare?mints= | $0.0015 / mint |
| /api/launches | $0.005 |
| /api/token/{mint} | $0.005 |
| /api/narrative/{mint} | $0.005 |
| /api/security/{mint} | $0.01 |
| /api/holders/{mint} | $0.01 |
| /api/history/{mint} | $0.01 |
| /api/wallet/{address} | $0.01 |
| /api/trending | $0.01 |
| /api/watchdog/pricing | free |
How to pay
- USDC, USDT or native SOL on Solana
- USDC on Base
- Standard x402 clients work without custom code — send
payload.transactionand a facilitator settles it - Settlement is not instant: retry a few times over 10–20 seconds while the payment is indexed
Connect as an MCP server
Every live-data endpoint and the guide catalog are also exposed as tools over Streamable HTTP MCP at https://saylorinnovations.com/mcp — add it as a tool server to Claude or any MCP-compatible agent and it can list, sample and buy directly, no custom integration code.
Discovery
- x402 manifest — every resource, price and payment requirement
- OpenAPI 3.1 spec — generate a client instead of reading docs
- Free search over the guide catalogue
https://saylorinnovations.com/mcp— connect as an MCP tool server
saylorinnovations.com. Use the URLs exactly as published in the manifest — that is the authoritative source.FAQ
What is x402?
x402 is an open protocol that reuses the HTTP 402 Payment Required status code as a real, machine-readable price quote. A server responds 402 with exact payment terms, the client pays on-chain, and retries the same request with proof of payment attached. No accounts, no API keys, no subscriptions — pay per call.
How do AI agents actually pay for data here?
An agent reads the 402 response, builds a small on-chain payment in USDC, USDT or SOL (or USDC on Base), and retries with an PAYMENT-SIGNATURE header. Settlement takes a few seconds to confirm, so a client should retry 3-4 times over 10-20 seconds rather than treating an initial "not found yet" as failure.
Does this work with Claude, ChatGPT, or other MCP-compatible agents?
Yes. /mcp is a Streamable HTTP MCP server. Connect it as a tool server to Claude or any MCP-compatible client and it exposes every live-data endpoint plus the guide catalog as callable tools, each with a free sample and its price.
Is there a free way to see what I’m buying first?
Yes. Every paid endpoint has a matching /api/sample/{kind} route that returns a real response for a fixed subject, free, so an agent can confirm the exact shape and data quality before spending anything.
What’s the difference between this and the Watchdog scanner?
Same underlying intelligence, two audiences. The scanner at /scanner/ is the free browser tool for humans — paste an address, get a score. This API sells that same intelligence per request to software: agents, bots, and anything querying many tokens programmatically.