Most local-model sizing advice is generic because the assistant has no idea what hardware it is talking about. This closes that gap with the smallest possible surface — two commands, read-only, nothing leaves the box. The project describes itself as beta, so treat the coverage as the starting set rather than the finished one.
A small local server that reports what your Mac actually is — hardware specifications, system configuration and current resource usage — so an assistant can answer machine-specific questions instead of guessing. It reads that state through `system_profiler`, and reads Python environment configuration through `conda`.
Your Mac's hardware specifications, system configuration and resource usage, made available to Claude Desktop or any other MCP client, plus the Python environment configuration that `conda` reports. It executes only those two commands, which is the whole scope of what it can see. It is aimed at people experimenting with AI and deep learning locally, where the answer to "will this model fit" depends on the actual machine.
A Mac, Python 3.8 or higher, and the UV package manager. No account and no key — everything it reads is local. Install by cloning the repository and pointing your client at it: `uv --directory /YOUR_PATH_TO/mcp-my-mac run -m mcp_server_my_mac`. The Python package is `mcp-my-mac` (0.1.0).
Build from source — clone the repository and build it, then point your client at the binary