FleetGrader

A MEDiAGATO product

Software you can drive with your AI.
That doesn't hand your data to the AI.

"Local-first" and "MCP-native" are the real terms for that, not a slogan. Here's exactly what each one means for FleetGrader, and why the combination is the actual point.

Local-first

Your fleet inventory has no server to leave.

FleetGrader reads the export file you already have and writes the report back to disk. There is no account, no sync, no upload step, and no database anywhere holding your clients' hostnames, serials, or site names — because none of that ever leaves your machine to begin with. If MEDiAGATO shut down tomorrow, every report FleetGrader has ever built for you would still open, and the app would still run. That's the actual test "local-first" has to pass — not a privacy policy, a fact about where the bytes are.

MCP-native

An AI assistant can drive it directly — through four narrow doors, not the front door.

MCP (Model Context Protocol) is how an AI assistant like Claude calls into a piece of software's own tools instead of guessing at a UI. FleetGrader ships a small MCP server, built into the same signed exe you already have — no separate download, no source access (off by default — FleetGrader.exe --mcp enable turns it on) — with exactly four tools, each one engineered to answer a question without ever handing back the thing the question was about:

fleet_health

Counts, not names

Client count, pattern count, how many issues are fire/soft/silent across every tracked site. Never a client's real name.

run_rollup

An outcome, not a path

Builds the real dashboard file, with real names in it, for you to open yourself afterward — the tool result handed to the assistant is the count of what it found, never a pointer to the file itself.

client_status

By ID, never by roster

Ask about one client you already know the ID for. A miss says "not found" — it never lists who else is tracked.

Why both, together

The assistant gets to help. Your clients' names don't get to leave.

This is the actual differentiator, not a feature checkbox: most "local-first" tools have nothing to say about AI at all, and most AI-drivable tools assume the assistant's own context is a safe place for your data to sit. FleetGrader's MCP surface is built the other way around — every tool result is a strict subset of what the engine actually knows, by construction, so an assistant with FleetGrader's tools enabled can tell you "your fleet health looks fine, one site has a new fire-tier issue worth a look" without your client's real hostname, serial, or company name ever landing in that assistant's transcript, its cloud provider's logs, or a screen-share. You get the help. The data stays put.

Questions

Ask if something here doesn't add up.

This page describes what's actually built and tested, not a roadmap. If you find a gap between this and the real behavior, that's a bug report.

[email protected]