Agent spending decisions and proof
Screwtape.ai decides whether an AI agent is allowed to spend money, and keeps the proof.
Screwtape.ai never holds your funds. With Payment Guard, a payment sent through Screwtape.ai only goes ahead when your rules allow it, and every decision comes with a signed receipt anyone can verify.
Agents pay per call automatically. No prepaid pack or pilot purchase is needed. Machine-readable pricing.
Tools agents can call
| Tool | What it does | Per call |
|---|---|---|
payment guardPOST /v1/payment-guard/authorize | Enforces your spend rules before an agent pays: a limit per payment, daily and period budgets, approved sellers and sites, and call rate limits. Allows with a signed clearance or blocks with reasons. Setting rules is free. | $0.005 |
trust scorePOST /v1/trust-score | For sellers: scores whether a paying agent looks risky before you accept its payment, from failed or refused payments, dispute patterns, speed and history across the Screwtape.ai seller network. Returns a 0 to 1 score with allow, review or block. | $0.005 |
spend booksGET /v1/spend-books/export | Exports what your agents spent on Screwtape.ai as bookkeeping rows by agent, seller, day and category, ready for QuickBooks or Xero, or as JSON. | Free |
judgePOST /v1/judge | Checks an agent output against your evidence and rules; returns pass, fail or retry. | $0.005 |
consensusPOST /v1/consensus | Compares agent answers and flags agreement patterns and false-consensus risk. | $0.005 |
duplicatePOST /v1/duplicate | Checks an invoice against previously recorded invoices and your supplied rules. | $0.005 |
budgetPOST /v1/budget | Tracks a run and its sub-agents against a shared cap; warns about spikes and repeated retries. | $0.002 |
receiptPOST /v1/receipt | Records a signed, hash-chained action receipt; verification is free. | $0.003 |
GET /v1/receipt/verify checks an action receipt for free. Use the verify page to paste an id. Every receipt is signed and hash-chained, and anyone can verify it for free.
Live usage
535 checks recorded · 62 action receipts issued. Counts come from Screwtape.ai's own database.
Listed in 9+ agent directories, including the Official MCP Registry, Glama and awesome-a2a. Featured in the LangChain docs.
What developers say
Merged — thanks @screwtapeai! Verified the Agent Card and that
message/sendworks against the live endpoint. Seller-payment safety checks (allow/review/block with signed receipts) fill a real gap for agent commerce — glad to have it in the payments section.
Start free
3 free calls per UTC day without a key, shared across the tools on HTTP and MCP by a private daily network identifier. Anonymous access is limited to 3 calls per minute and 2,000 calls per day across the service. A free API key adds 10 free calls per UTC day, shared across the tools. Create it with POST /v1/signup; email is optional. The free key needs a small proof of work to limit abuse. Anonymous ledger and receipt-chain identity changes each UTC day; use a key for a persistent identity. No raw network address is stored by the anonymous counter.
For an extra fixed sample without using that allowance, use the MCP try_check tool or POST /v1/try-check. It checks a fixed sample invoice with deterministic rules; it does not run your own workload.
Connect your agent
The API documentation, spend guardrail guides, and MCP connection guides show how to connect.
Add this MCP URL to your AI:
{
"mcpServers": {
"screwtape": {
"url": "https://screwtape.ai/mcp",
"headers": {
"Authorization": "Bearer ${env:SCREWTAPE_API_KEY}"
}
}
}
}MCP server: https://screwtape.ai/mcp. Tools include payment_guard_authorize, trust_score, judge, consensus, check_duplicate_invoice, budget_ledger and action_receipt. What is new in 1.7. Optional header X-Caller-Label tags usage per fleet (letters, digits, . _ : -, max 64).
