AgentLFG
One searchable inventory for every MCP server and skill scattered across your AI dev tools.
The question I was chasing
How many MCP servers and skills do I actually have installed, and where?
Every AI dev tool keeps its own config. Claude, Cursor, Codex, LM Studio, VS Code: each one stores its servers and skills somewhere different, in its own format. I'd lost track of what was wired up where. The question was whether you could see the whole agentic surface of your machine in one place.
Why it exists
MCP went from novelty to everywhere in months. The cost of that speed is sprawl: duplicate servers, dead entries pointing at tools you uninstalled, the same skill configured three different ways across three apps. None of it is visible until something breaks.
The constraints
This tool scans a developer's machine and reads config that can contain API keys and tokens. That set one hard rule before any feature: read-only by default, zero telemetry. The moment a tool like this looks like it might exfiltrate a credential, no developer will run it. It also had to be fast and native across half a dozen different config schemas.
The decisions that mattered
Read-only by default was the decision everything else hung off: trust first, features second.
Tauri and Rust over Electron, because a tool that touches the filesystem and scans configs should be small, fast and native, not a browser carrying a runtime.
The genuinely hard part wasn't the UI; it was normalisation. Six tools describe MCP servers in six different shapes (stdio, HTTP, SSE, varying field names). Collapsing them into one model you can actually search and compare was most of the work. Installing new MCPs straight from GitHub was what turned it from a diagnostic into something you reach for.
What it is
A native macOS app that scans your Mac for every skill and MCP server across Claude, Claude Desktop, Cursor, Codex, LM Studio and VS Code, and surfaces them in one searchable inventory. Audit duplicates, spot orphans, inspect transports, install new MCPs from GitHub. Read-only by default, zero telemetry, free.
Built with: SvelteKit, TypeScript, Tauri (Rust), Claude Code, Cursor
Where it landed
It shipped free, and it's the tool I use to keep my own agentic setup from rotting. If I took it further, the obvious next step is drift detection across machines: telling you what changed since last week, not just what's there now.
Part of the Rolling Waves work archive.