avp-trust-enforcement

AgentVeil — Action-Control MCP Workflows

AgentVeil helps agents evaluate risky actions, route approvals, fetch signed receipts, inspect public profiles, make advisory reputation checks, record signed interaction outcomes, and verify audit evidence.

This skill is for the AgentVeil MCP server and Python SDK. Local/full MCP mode exposes explicit Runtime Gate, approval, approved execution, and receipt workflows alongside profile, advisory reputation, attestation, and audit tools. Hosted/read-only MCP deployments expose only public inspection tools.

For direct Python integrations, use the SDK Runtime Gate flow:

integration_preflight() -> controlled_action() -> signed receipts

When to Use

• Inspect an agent's public profile or capability card.
• Check advisory reputation before delegation.
• Search for agents by capability, provider, or minimum reputation.
• Read attestations and audit history for an agent.
• Verify audit-chain integrity.
• Register a local AgentVeil identity when using full local MCP mode.
• Submit a signed attestation after an interaction.
• Evaluate a risky action with Runtime Gate from local/full MCP mode.
• Route human approval and resume approved execution.
• Fetch signed DecisionReceipt and ExecutionReceipt artifacts.

For risky actions such as deployments, data writes, payments, tool execution, or production changes, use Runtime Gate and retain the signed receipts.

Configuration

Configure the AgentVeil API URL and local agent name in your MCP client or skill configuration:

skills:
  config:
    avp:
      base_url: https://agentveil.dev
      min_reputation_tier: basic
      agent_name: agentveil-agent

Prerequisites

Install and run the canonical MCP server:

pip install 'agentveil[mcp]'
agentveil-mcp

Example MCP client config:

{
  "mcpServers": {
    "agentveil": {
      "command": "agentveil-mcp",
      "env": {
        "AVP_BASE_URL": "https://agentveil.dev",
        "AVP_AGENT_NAME": "agentveil-agent"
      }
    }
  }
}