Agent Verifier MCP
goodmeta/agent-verifier-mcp
Enforce budget limits para agentes IA em qualquer protocolo de pagamento
Ver no GitHubServidor MCP que implementa o Budget Authority Protocol, fornecendo limites de gastos para agentes IA. Controla spending em tempo real com hold-settle pattern, funcionando de forma agnóstica ao método de pagamento (x402, cartão, MPP, transferência bancária).
Instalação
npx convez-ai goodmeta/agent-verifier-mcp1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Limitar gastos de agentes IA em chamadas a APIs pagas (Exa, Firecrawl, Nansen)
- Implementar controle de orçamento multi-serviço para prevenir overspend
- Rastrear gastos de agentes com hold-settle-refund workflow
Veredito

MCP server bem-arquitetado que resolve um problema real: agentes IA sem visibilidade de gastos totais entre múltiplas APIs. Implementação limpa com duas opções de deploy (local/remoto) e integração direta ao Claude, sem overhead de setup.
README
A tradução para o português deste README ainda está na fila.
Agent Verifier MCP
MCP server implementing the Budget Authority Protocol. Spending limits for AI agents. Two ways to connect:
Option A: Local (npx)
{
"mcpServers": {
"budget": {
"command": "npx",
"args": ["@goodmeta/agent-verifier-mcp"],
"env": { "VERIFIER_API_KEY": "gm_..." }
}
}
}
Option B: Remote (no install)
{
"mcpServers": {
"budget": {
"type": "streamable-http",
"url": "https://verifier.goodmeta.co/mcp",
"headers": { "Authorization": "Bearer gm_..." }
}
}
}
Then ask your agent:
"Search Exa for agent payments research. Budget: $1."
The agent calls create_budget, check_budget before each payment, and settle after. If the budget runs out, the next payment is denied.
Tools
| Tool | What it does |
|---|---|
create_budget | Set a spending limit ($10, 24 hours) |
check_budget | Check if a purchase is allowed (places a hold) |
settle | Confirm hold after payment, or release if payment failed |
release | Return a pre-commit hold to the budget (cancel before it settles) |
refund | Reverse a settled payment, fully or partially |
get_budget | Query remaining budget and history |
query_reservation | Look up a single hold's state (held, settled, released, refunded) |
How it works
Agent wants to buy something ($0.50)
│
├── check_budget(budget_id, 50, "vendor")
│ → "Approved. Hold ID: hold_abc. Remaining: $9.50"
│
├── [agent makes the payment via x402/card/MPP/any rail]
│
└── settle(hold_id, success=true)
→ "Hold confirmed."
Budget enforcement is rail-agnostic. Works with x402, credit cards, MPP, bank transfers — the verifier tracks the budget, the payment method doesn't matter.
Setup
Get a free API key:
curl -X POST https://verifier.goodmeta.co/setup/merchants \
-H "Content-Type: application/json" \
-d '{"names":["My Agent"]}'
Add the key to your MCP config. Done.
Works with
- Claude Code
- Any MCP-compatible client (Cursor, Windsurf, Codex, OpenClaw)
- agent-verifier npm package for programmatic use
- demo-agent for a full working example with real x402 payments
Why
Five protocols let AI agents spend money (AP2, ACP, x402, MPP, UCP). None of them track what the agent spent across services. An agent calling Exa + Firecrawl + Nansen can overspend because each service approves independently. This MCP server is the missing budget layer.