Avowex MCP Server

Give any AI agent a conscience and a paper trail. The Avowex MCP server lets a model (Claude, Cursor, Cline, …) pause a risky action, route it to a human for approval, and record the whole thing in a tamper-evident audit trail — the evidence you need for the EU AI Act's human-oversight and logging requirements and for SOC 2.

Free tier: 500 resolved escalations / month. Get a key at the Avowex console.

Install

Add to your MCP client config (Claude Desktop, Cursor, Cline all share this shape):

{
  "mcpServers": {
    "avowex": {
      "command": "npx",
      "args": ["-y", "@avowex/mcp"],
      "env": {
        "AVOWEX_API_BASE": "https://api.avowex.com",
        "AVOWEX_API_KEY": "avx_live_your_key"
      }
    }
  }
}

Tools

| Tool | What it does | |------|--------------| | avowex_escalate | Wrap a risky action in a human-in-the-loop check. Pass context.agent to track each agent. | | avowex_get_decision | Fetch the human decision (approve / reject / edit). | | avowex_resume | Resume the agent after applying the decision. | | avowex_verify | Check a proposed action against policy (no human needed). | | avowex_get_audit | Verify the tamper-evident audit chain — your compliance evidence. |

Typical flow

1. Your agent is about to do something consequential (issue a refund, send a blast, move money).
2. It calls avowex_escalate with the action and context.
3. A human approves / rejects / edits in the Avowex reviewer app or via Slack / SMS / email.
4. Your agent calls avowex_get_decision, applies it, then avowex_resume.
5. Every step is hash-chained into an audit trail you can verify with avowex_get_audit.

Config

| Env | Default | Notes | |-----|---------|-------| | AVOWEX_API_BASE | https://api.avowex.com | Point at http://localhost:8000 for local dev. | | AVOWEX_API_KEY | — | Required to make real calls. Free tier included. |

Develop

npm install
npm run dev      # tsx src/index.ts
npm run build    # -> dist/index.js

License: Apache-2.0