Coming soon to the Mac App Store

AIAgentGate

See what your AI agents are configured to reach.

Your coding agents and their MCP servers are configured by files scattered across your home folder. They accumulate: a server installed once and forgotten, a filesystem root far wider than the project needed, an API key pasted in plain text, a package that fetches a new version on every run. AIAgentGate reads that configuration and tells you what it adds up to — with the evidence for every finding.

Requires macOS 14 or later · one-time purchase, no subscription

Nobody set out to grant all of this

No single decision here was careless. Each server was added for a reason, each root was widened to make something work, each key was pasted to get unblocked. The problem is that nothing ever reads them back to you together.

So the honest answer to “what can the agents on this machine get at?” is usually that nobody knows — not because anyone was reckless, but because the answer is spread across a dozen files nobody opens.

Inventory

Every server, and what it actually declares

AIAgentGate reads the known agent and MCP configuration locations in your home folder — each one granted individually, through a standard macOS open panel — and puts what it finds into one list. For each server that means the command it runs, its parsed arguments, the transport it uses, the names of the environment variables it expects, the filesystem roots it declares, and the remote endpoints it names.

Names, not values

Environment variables are inventoried by name. The point is to show you which secrets a server expects, without becoming another place they are written down.

Arguments parsed, not just printed

A command line is read as arguments, so a filesystem root buried three flags deep is surfaced as a root rather than left inside a string.

Files it could not parse

Anything unreadable is reported as exactly that. A file skipped in silence would be the one place an audit could quietly lie to you.

Findings

What it flags, and the evidence for it

Each finding names the file and the configuration that produced it, so you can judge it yourself rather than take the app's word for it.

Plaintext credentials — secrets written directly into a configuration file rather than referenced.

Over-broad filesystem roots — a server granted far more of your disk than the work needs.

Shell and interpreter invocations — servers launched through a shell or an interpreter, where the effective command is not what it appears.

Floating package tags — entries that fetch whatever the latest version happens to be, so what runs tomorrow is not what you reviewed today.

Remote endpoints with no declared auth — a configured endpoint with nothing in the configuration saying how it authenticates.

Duplicate tool names — the same tool name exposed by two servers, where which one answers is not obvious.

Unparseable files — configuration the audit could not read, reported rather than skipped.

Change over time

Baselines, diffs, and a fix you can undo

Save a baseline once you have reviewed things, and later runs tell you what changed rather than making you re-read the whole inventory. A new server, a widened root or a newly floating tag shows up as a difference from the state you approved.

When a finding has an obvious remedy, AIAgentGate can apply it: it backs the file up first, validates the result, and rolls back if the result does not hold up. You can also export a redacted report for a security review — no absolute paths, no secret values — so the artefact you hand someone is not itself a disclosure.

It audits. It does not guard.

AIAgentGate inspects configuration and reports findings. It does not sit between an agent and your files, and it cannot prevent an agent from doing anything. Nothing it reports is enforced by it.

It also reads configuration rather than watching an agent run, so what it tells you is what your setup declares or permits — not a record of what any agent actually did. It never starts an MCP server and never runs a command it finds in a configuration file.

Coming soon

AIAgentGate is being prepared for the Mac App Store. The documentation is already written if you want to see exactly how it works first.

Read the documentation