agent-envelope-mcp

The neutral authority layer for agent runtimes, as an MCP server.

Any MCP client — Claude, an OpenAI agent, LangChain, CrewAI, a custom runtime — can call these tools to check and issue agent authority without building its own policy engine, audit log, or verification stack. AgentEnvelope is owned by no framework, so every framework can embed it.

Run

npx agent-envelope-mcp

It speaks MCP over stdio. No API key is needed to start, or to use the sovereign verifier — only the vault-governed tools require AE_API_KEY.

Wire it into an MCP client

Point your client at the command and pass the vault-issued key in the environment (only needed for the custody tools):

{
  "mcpServers": {
    "agent-envelope": {
      "command": "npx",
      "args": ["-y", "agent-envelope-mcp"],
      "env": { "AE_API_KEY": "your-vault-issued-key" }
    }
  }
}

Tools

• ae_verify_sovereign — verify a signed message against a known agent address. Pure crypto: no vault, no key, no network. Verification is always free.
• ae_get_agent — fetch the vault-registered public record for an agent id.
• ae_verify_action — verify a signed action against the vault-held record.
• ae_mint — mint a capability through the vault from a MintDelegate and a signed MintRequest, returning a mint receipt. This is a governed action.

The sovereign verifier needs no vault and no key. The vault-gated tools are the governed surface a framework offloads rather than rebuilds.

Programmatic use

import { createServer } from 'agent-envelope-mcp';
// mount createServer() on your own MCP transport

Environment

The server is fail-closed: vault-gated tools return an error result when AE_API_KEY is absent rather than exposing an unauthenticated surface.

License

Apache-2.0 — see NOTICE for attribution.