Potencia tu agente
El comando propio del publicador. No es un endpoint de PlugMCP.
Host a copy with our keys and logs.
npx -y @deepfirstsearch/[email protected]Claude Code no tiene enlace de instalación para un comando local. Copia el comando de abajo y ejecútalo en una terminal.
claude mcp add deepfirstsearch-agent-pay-mcp -- npx -y @deepfirstsearch/agent-pay-mcp@0.4.0Ejecútalo en una terminal.
{ "mcpServers": { "deepfirstsearch-agent-pay-mcp": { "command": "npx", "args": [ "-y", "@deepfirstsearch/[email protected]" ] } }}claude_desktop_config.json
¿URL remota o un comando en tu máquina? En qué se diferencian las dos instalaciones
Ejecutar en PlugMCP
A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.
AbrirDel publicador
Safe rails for the agent economy. Search deep. Reveal only what you choose.
AI agents now pay for things over x402. Every one of those payments is public, and any agent can be talked into paying the wrong party. Deep First Search makes the model propose payments while the owner's signed policy .
MCP server: any MCP agent (Claude, Cursor...) pays x402 APIs inside an owner-signed, on-chain budget
No keys, no wallet, no chain: a local x402 merchant plays honest and hostile, and you watch the agent pay for data and refuse four attacks.
npx @deepfirstsearch/agent-pay demo
Already building an x402 client? See what it signs when the server lies — a swapped payee, fake decimals, no price cap — in one command:
npx @deepfirstsearch/x402-scan client @x402/fetch
x402-scan is a free, zero-dependency scanner (MIT); the findings explain what we keep seeing across published clients. The fix for everything it flags is the policy layer below.
The integrations take the SDK as a peer dependency, so your app always uses a single copy of it. Step-by-step guides for 40 x402 merchants, each tested with a real payment on Base mainnet, are in docs/integrations.
Full developer guide: deepfirstsearch.com/developers · building a wallet, framework or platform? Partners.
@deepfirstsearch/agent-pay-mcp is a stdio MCP server. The model gets three tools and nothing that can move money on its own terms:
{ "mcpServers": { "agent-pay": {
"command": "npx", "args": ["-y", "@deepfirstsearch/agent-pay-mcp", "/path/config.json"],
"env": { "AGENT_PAY_AGENT_KEY": "0x…", "AGENT_PAY_BURNER_SEED": "0x…" } } } }
Payees, prices and limits come from the config file and the owner-signed on-chain budget, never from the model or a 402 response, so a prompt-injected agent can't overspend or pay an attacker. Setup, config reference and a 5-minute Base Sepolia demo: integrations/mcp.
Status: pre-launch. Live on Base Sepolia, with a real x402 payment settled (see deployments); a capped beta on Base mainnet since 2026-10-06 (small amounts, our own funds first, then invited design partners). The TypeScript SDK has had an independent external review (findings fixed); the on-chain contracts are not yet audited — a professional audit comes before the public launch. No token exists.
# Contracts (needs Foundry: curl -L https://foundry.paradigm.xyz | bash && foundryup)
git submodule update --init --recursive
cd contracts && forge test
# SDK
cd ../sdk && npm ci && npm test && npm run demo
# Website
cd ../web && npm ci && npm run dev
$DEPTH.There's no telemetry in the packages, so we only know who uses them if you say so. If your agent pays with Agent Safe, even once on testnet, tell us in this discussion: what it pays for, what got refused that shouldn't have, and what's missing. We'll test your sellers live and write a verified guide for any that's missing.
See SECURITY.md to report vulnerabilities.
Built by Nicolas Tursi. General contact: [email protected]. Security reports: see SECURITY.md or write to [email protected].
MIT (contracts and SDK).
MCP server: any MCP agent (Claude, Cursor...) pays x402 APIs inside an owner-signed, on-chain budget
{ "mcpServers": { "deepfirstsearch-agent-pay-mcp": { "command": "npx", "args": [ "-y", "@deepfirstsearch/[email protected]" ] } }}Herramientas
Aún no hay lista de herramientas. Los servidores remotos la reciben tras la primera comprobación de estado.
Preguntas frecuentes
Copia en esta página la configuración stdio para Cursor, Claude Desktop, Claude Code o VS Code. Usa la instalación del publicador, no un proxy de PlugMCP.
Este registro no indica ninguna variable de entorno obligatoria.
@deepfirstsearch/agent Pay se ejecuta localmente por stdio.
Última comprobación: 2026-11-10. MCP server: any MCP agent (Claude, Cursor...) pays x402 APIs inside an owner-signed, on-chain budget