veto-cli 0.3.0

Veto CLI — one-command MCP setup for AI agent payment authorization.

Veto CLI

One-command setup for the Veto authorization layer — protect every payment your AI agent makes.

Veto is the policy and approval layer for AI agents that take real actions: x402 payments, Stripe Issuing transactions, on-chain transfers. The Veto CLI auto-configures Veto for any MCP-compatible client (Claude Desktop, Claude Code, Cursor, Zed, Continue) so your agent calls Veto before every transaction — and the transaction is allowed, denied, or escalated for human approval based on policies you define.

Install

pip install veto-cli

Requires Python 3.9+. No third-party dependencies — stdlib only.

Quickstart — three commands

# 1. Install
pip install veto-cli

# 2. Register an account from the terminal (no website, no form)
veto register --email [e-mail] --preset dev
# → ✓ Welcome to Veto. API key + default agent saved locally.

# 3. Ask Veto whether an action is allowed
veto authorize --amount 0.05 --merchant api.anthropic.com --action payment
# → APPROVED / DENIED / ESCALATED. Exit code 0/1/2.

That's it. Three commands, zero web pages, working agent-payment gatekeeping.

Policy presets

veto register applies a policy preset so your agent has sensible limits from the first authorize call. Pick one with --preset:

Preset	For	Defaults
personal (default)	General-purpose agent	$500/tx, $2k/day, blocks gambling/mixers/adult
inference	AI API calls	$5/tx, allowlists Anthropic/OpenAI/Replicate/etc.
x402-micropay	x402 testing	$1/tx, Base chain only, auto-approve <$0.10
ad-spend	Meta/Google ads	$1k/tx, escalate >$1k
dev	Dogfooding/testing	$500/tx, no merchant restrictions

Override the agent's mission with --mission "...". Override the agent name with --agent-name "...". Edit policies later with the veto policies commands (coming in 0.4.0).

Headline command — veto authorize

veto authorize \
  --amount 0.05 \
  --merchant api.anthropic.com \
  --action payment

# → 0 if approved, 1 if denied, 2 if escalated, 3 on error.

After veto register, the default agent UUID is saved locally — you don't need to pass --agent on every call.

JSON output for piping into other tools:

veto authorize --agent ... --amount 0.05 --merchant ... --action payment --json

Read input from stdin:

echo '{"agent_id":"...","amount":0.05,"merchant":"...","action":"payment"}' | veto authorize -

Why this matters

veto authorize returns the decision — approve, deny, or escalate — without any side effect. Your agent stays in control of the actual payment / signing / API call; Veto just gatekeeps. That's Mode 1 (decision API).

veto test and veto init-installed MCP integration also support Mode 2 (Veto creates a Stripe-issued virtual card from your authorized request), but Mode 1 is the headline use case for any agent that already has its own wallet, card, or rails.

Commands

Command	What it does
veto authorize	Ask Veto whether an agent action is allowed (returns approve / deny / escalate). Headline command.
veto init	Auto-detect MCP clients on your machine and add Veto to each one's config
veto status [agent_id]	Show your agent's current reputation tier and recent decision history
veto test [agent_id]	Fire a synthetic Mode-2 test transaction (creates a real Stripe-issued virtual card)
veto list	List installed MCP clients and Veto integration status
veto uninstall	Remove Veto from MCP client configs (does not delete your account)
veto mcp	Run the Veto MCP server in foreground (used by MCP clients)

What Veto evaluates on every authorize call

Each transaction passes through an 8-step pipeline before approval:

1. Pre-checks — agent suspended? amount sane?
2. Policy enforcement — per-tx limit, daily/monthly caps, merchant allowlist/blocklist
3. Prompt injection detection — 40 regex patterns over the action description
4. Merchant fraud screening — known-fraud database, typosquatting (SequenceMatcher), suspicious TLDs
5. Intent verification — does the action match the agent's stated purpose?
6. Anomaly detection — amount spike (>3× rolling avg), velocity, merchant diversity
7. LLM final verdict — Claude Sonnet reviews the case
8. Reputation weighting — agent trust tier modulates final risk score

Output: approve | deny | escalate (with risk score 0.0–1.0 and a human-readable reason).

Configuration

The CLI stores state in ~/.veto/config.json (mode 0o600). It contains your API key and known agent IDs. No transaction data is stored locally.

By default the CLI talks to https://veto-ai.com. To point at a self-hosted Veto:

veto init --api-key XXX --base-url https://veto.your-company.com

Links

• Sign up: https://veto-ai.com
• Docs: https://veto-ai.com/docs
• Discord: https://discord.gg/veto-ai
• GitHub: https://github.com/veto-protocol

License

MIT. See LICENSE.