MCP-ready ┬À Deterministic ┬À Auditable ┬À Pay with USDT

Your agent has a wallet. MandateGuard decides what it can spend.

Autonomous agents get drained every day ÔÇö $45M lost in a single January attack. MandateGuard is the pre-action enforcement layer: budgets, allowlists, rate limits, and signed mandates evaluated with no LLM in the decision path. Mount it via MCP in minutes.

Buy Pro ÔÇö $149/mo GitHub
$45M+documented agent losses
0 msLLM latency in decisions
38tests, replay-verifiable
v1.0.0production-ready

The problem: excessive agency

OWASP names it LLM08. Every agent with a wallet is one prompt injection away from being drained. Standards (Google AP2, Coinbase x402, ERC-8004) define the paperwork ÔÇö nobody ships the enforcement.

Drain attacks

Prompt injection makes the agent "refund the customer" ÔÇö to an attacker wallet. MandateGuard blocks by allowlist, budget, and per-call limits before the tool runs.

Memory poisoning

Compromised context steers the agent to bulk spend. Signed, time-boxed mandates cap the blast radius; nothing exceeds the issuer's explicit scope.

No audit trail

When a drain happens, you need proof. Every decision lands in a tamper-evident SHA-256 ledger you can verify in O(n).

How it works

One deterministic engine. No model sampling. Same inputs  same verdict, always.

Agent intent ÔöÇÔöÇÔû authorize(intent) ÔöÇÔöÇÔû PolicyEngine                                          Ôöé  scope ┬À allowlist ┬À denylist                                          Ôöé  budget ┬À rate limit ┬À mandate                                          Ôû╝                                     APPROVED / DENIED / REQUIRES_APPROVAL                                          Ôû╝                              append-only SHA-256 ledger (audit)

Per-actor scopes

Allowed tools, destinations, per-call max, currency, per-window call limits.

Signed mandates

Ed25519, nonce-bound, time-boxed authorizations in the AP2 / x402 / ERC-8004 pattern. The agent cannot widen its own scope.

MCP server

Mount as a guardrail in Claude, Cursor, or any harness. Pro features (RBAC, revocation, persistence, webhooks) gate via signed licenses.

Pricing

MIT core, forever. Pay only for what operations need: revocation, persistence, RBAC, webhooks, and audit notary. Pay in USDT ÔÇö Solana or Ethereum.

Community
Free
MIT license
  • Deterministic engine
  • Scopes + global guards
  • Signed mandates
  • Tamper-evident ledger
  • Basic MCP server
Try live demo Get started
Most popular
Pro
149 USDT
per month ┬À 5 seats
  • Everything in Community
  • Mandate revocation registry
  • Policy + state persistence
  • RBAC & multi-tenant
  • Webhook alerts
  • Signed Pro license
Buy Pro
Enterprise
Custom
volume & SLA
  • Everything in Pro
  • SSO / SAML
  • Audit notary (external hash)
  • Support SLA
  • On-prem deployment
Contact sales

Buy Pro

Send USDT to one of the wallets below and email your order id + tx hash to . Your signed Pro license is issued within 24h.

Option A ÔÇö Solana (USDT-SPL)

Faster and cheaper fees. Amount: 149 USDT/mo.

3fZSMAyCEMhZwWiynbJDjoYNUT97aiV9BLzoUNroEMAz

Option B ÔÇö Ethereum (USDT-ERC20)

Amount: 149 USDT/mo.

0x4Ed4D0750453C027FA8398067d5af980Bcc9B6eD

Include your email or order id in the transfer memo. After sending, email your tx hash so we can verify on-chain and deliver your license.

Install

# from source (repo, works today)git clone https://github.com/ezequiellich44-cmd/MandateGuard.gitcd MandateGuardpip install -e ".[mcp]"mandateguard-mcp# or mount the MCP bundle from the official registry:io.github.ezequiellich44-cmd/mandateguard

Python 3.10. Core decision path is stdlib + cryptography for mandates. Full docs on .

Ship your agent. Keep your wallet.

Questions or volume pricing?

Star on GitHub