From agentpay
Find, judge, and pay for the best paid x402 tool/API for a need, within a budget. Use when the agent needs a capability behind a paid API and must choose a provider: "which x402 service / paid API for X", "find a tool for X under $Y", comparing providers, avoiding overpriced or fake/stub endpoints, or routing agent spend wisely. Picks "the cheapest tool that's real and actually used" — never cheapest, never a stub.
How this skill is triggered — by the user, by Claude, or both
Slash command
/agentpay:agentpay-routeThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
When the agent needs a paid capability it doesn't already have, use AgentPay to **discover
When the agent needs a paid capability it doesn't already have, use AgentPay to discover candidate x402 tools across the marketplace, judge which one is real and actually used (not the cheapest, not a keyword-stuffed stub), pay the best one under a budget, and keep a receipt. This is buyer-side economic intelligence: the agent shops the marketplace instead of grabbing the first or cheapest result.
Option A — MCP route tool (any MCP-capable agent, no setup):
If your agent runtime has the AgentPay MCP connected (npx @romudille/agentpay-mcp), call
the route tool directly — no Python, no CLI, no repo:
{ "tool": "route", "arguments": { "need": "funding rates", "budget": 0.01 } }
Returns a ranked candidate list + recommendation + ready_to_pay details (provider URL +
x402 accepts entry) as JSON. Advise-only, keyless, works in any MCP runtime.
Option B — bundled CLI router (Claude Code / shell, pure stdlib, no setup):
agentpay-route "<what you need>" --budget <max USDC, e.g. 0.01>
It queries Coinbase Bazaar, drops stubs (no schema, keyword-stuffed, or factory clones — one wallet behind many fake names), ranks survivors by real usage (unique payers > raw calls, plus recency), enforces the budget, and prints a ranked table + a recommendation with the provider URL and why.
2. Apply taste. The router supplies price / quality / legitimacy; the agent makes the final capability call — does this tool actually return the field the task needs? Pick from the ranked list (usually the ★ recommendation).
3. Pay the provider directly, capped + receipted (peer-to-peer). Use the AgentPay SDK so the spend stays under a hard cap and every call produces a verifiable receipt + ledger:
pip install agentpay-x402
from agentpay import Session, AgentWallet
s = Session(AgentWallet(secret_key="S...", base_key="0x..."), max_spend="0.05")
r = s.call("<chosen-provider-url>", {...}) # pays the provider directly via x402
print(s.spending_summary()) # receipt + running ledger
{} is worse than a $0.001 tool
with 25 real payers. The router encodes this — trust its ranking over raw price.Home: https://agentpay.tools · routing is advise-by-default (you choose and pay).
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.
npx claudepluginhub romudille-bit/agentpay --plugin agentpay