For AI Agents
PayAPI Market is the easiest way for an autonomous agent to find and pay for real-world data and tools. One MCP endpoint, dozens of APIs, USDC settlement on Base.
Last updated: 2026-04-11
Connecting from Claude Desktop
Add this to your claude_desktop_config.json:
{
"mcpServers": {
"payapi": {
"url": "https://web-production-18a32.up.railway.app/mcp/sse"
}
}
}
Restart Claude Desktop. You'll see every PayAPI Market tool in the tool list.
Connecting from Cursor / Continue / custom agents
Any MCP-compatible client works. Point it at the same SSE endpoint above.
Paying for calls
Your client needs to:
- Have a USDC wallet on Base with a small balance
- Handle HTTP 402 responses by signing and broadcasting the requested payment
- Retry the request with proof of payment
The official x402 client libraries (TypeScript, Python, Rust) handle this automatically. Total per-call latency including payment is ~500-800ms.
Why this matters for agents
Subscription APIs are unusable for agents. You don't know in advance whether you'll need 5 calls or 50,000, you can't sign up for a $99/month plan when each user session might justify $0.02 of API spend, and you can't manage 50 different API keys across 50 different vendors. PayAPI Market collapses all of that into a single MCP endpoint and a USDC wallet.
Frequently asked questions
Do I really not need an API key?
Correct. The whole point of x402 is that authentication and authorisation are bundled into a single payment step. The server returns HTTP 402 with payment requirements, your agent signs and broadcasts the USDC transfer, the request goes through.
What does a typical call cost?
Most APIs on PayAPI Market are priced between $0.0005 and $0.01 per request. A typical agent task uses 2–20 API calls and costs less than a cent in total.
Which wallet should my agent use?
Any USDC-on-Base wallet. The official x402 client libraries handle key management, signing, and broadcasting; you just need to fund a Base address with USDC.
How long does payment add to the request?
Approximately 200–400 ms on top of normal latency, for a total per-call latency of 500–800 ms. Payment is verified in parallel with the request handler when possible.
Can I set a per-call spending cap?
Yes, in your MCP client config. Set max_cost_per_call (in USDC) and the client refuses requests that exceed it.