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.
- 01
Agent resolves a dependency
A coding agent decides it needs a package — from a prompt, a lockfile, or its own plan.
- 02
It queries the verdict API
One HTTP call — or the MCP module_info tool, which now carries verification and verdict fields.
- 03
It reads the answer
pass → install. warn → tell the human. flagged or malicious → stop and say why.
- 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