Damat registry

Built for AI agents

Your agent asks before it installs.

AI agents install dependencies at machine speed — and typosquats, hijacked maintainers, and poisoned versions are exactly the attacks they cannot eyeball. The registry gives agents a question to ask first: is this package legit?

One endpoint

GET /packages/:name/:version/verdict

A structured answer written for machine consumption — status, score, reasons, and a plain-language summary an LLM can reason over.

{
  "status": "pass",
  "score": 96,
  "reasons": [
    "owner verified (org: lodash)",
    "provenance attestation valid",
    "no static findings",
    "AI scan: no suspicious behavior"
  ],
  "summary": "lodash@4.17.21 is published by its verified
   owner, carries valid provenance, and no scan raised
   findings. Safe to install.",
  "checkedAt": "2026-07-09T18:12:04Z",
  "scans": { "static": "done", "ai": "done" }
}

The loop

Four steps, no trust falls.

  1. 01

    Agent resolves a dependency

    A coding agent decides it needs a package — from a prompt, a lockfile, or its own plan.

  2. 02

    It queries the verdict API

    One HTTP call — or the MCP module_info tool, which now carries verification and verdict fields.

  3. 03

    It reads the answer

    pass → install. warn → tell the human. flagged or malicious → stop and say why.

  4. 04

    The gate backs it up

    Even if an agent skips the question, the install itself passes through the same policy gate.

MCP-native

The registry is already in your agent's toolbox.

The Damat MCP server exposes search_modules, module_info, and add_module against the hosted registry — with verification status and trust verdicts in the responses, so the safety check happens inside the tool call.

Get started

Put a gate in front of your installs.

registry=https://registry.damatjs.com/api/