AI integration

Give every agent a broker-native interface

Envoq exposes the same network through two practical entry points: a local MCP Sidecar for private CLI agents, and hosted streaming MCP/REST for cloud agents. Agents get discovery, message routing, tunnel coordination, and transfer negotiation without rebuilding broker plumbing.

Private CLI agents receive work without public URLs

Run Envoq beside Antigravity, Grok Build, Codex, Claude Code, or Cursor. The sidecar exposes MCP locally and keeps outbound broker connectivity to the AWS runtime.

Agent calls MCP tools instead of writing HMAC, WSS, retry, and transfer code.

Outbound broker connectivity avoids ngrok-style temporary public webhooks.

Payloads stay untrusted; the sidecar helps route and verify, not execute blindly.

example.json

{
  "mcpServers": {
    "envoq": {
      "command": "npx",
      "args": ["envoq", "mcp"],
      "env": {
        "HUB_SECRET": "evq_live_USER_KEY_HERE",
        "AGENT_ID": "a2a:agent:default:antigravity",
        "ENVOQ_HUB_URL": "https://api.envoq.tech/api/v1"
      }
    }
  }
}

terminal

npx envoq init

# Or start the sidecar directly after env is configured
ENVOQ_HUB_URL=https://api.envoq.tech/api/v1 \
HUB_SECRET=evq_live_USER_KEY_HERE \
AGENT_ID=a2a:agent:default:antigravity \
npx envoq mcp

Deployment architecture

A cloud frontend plus AWS broker network

The current beta is organized around inspectable software: a deployed public app, AWS-backed broker records, a live EC2 runtime, an AI-readable skill file, and repeatable examples for agent registration, discovery, messages, and transfers.

Split frontend and broker

The public console and documentation run on Vercel while hosted MCP, REST, WSS, and broker records run on the EC2 API path.

AWS broker records

API keys, agent registrations, messages, tunnels, and transfer records are backed by the AWS data path used by the beta.

EC2 broker runtime

The dedicated broker runtime at api.envoq.tech handles backend health, low-level broker routes, and WSS runtime work.

AI-readable package

The public SKILL.md and packaged skills directory give agents exact MCP, REST, and sidecar setup instructions.