Do math on this machine with NumPy and plots

Who it is for

For people who can already run Python 3.12+ and uv on this machine and want an assistant to do numeric work, symbolic math, or plots. Fermat MCP is a FastMCP service that puts numeric, symbolic, and plotting together. Docs align it with NumPy, SymPy, and Matplotlib.

Hard prerequisite: Python 3.12 or newer. Install notes point to https://www.python.org/downloads/ and starting server.py with uv from a cloned repo. The official config's --directory points at /home/ty/Repositories/fermat-mcp on the author's machine; you must replace it with your own clone path. Do not copy it as-is.

Not for people who cannot clone the repo, or who only want one remote install command. The Smithery gemini install is another path, not the main one. Also not for people who treat plot output as a guaranteed on-disk path — current docs do not say where images are written.

First try in 3 minutes

Official local start is uv, running server.py inside the repo directory. Config lives in the host MCP entry: command is uv, args are --directory, your clone path, run, server.py. First confirm this machine is at least Python 3.12 and uv is installed.

On the first try do a linear-algebra calculation, for example det or dot. Docs list under Linear Algebra: dot, matmul, inv, det, eig, solve, svd. Pick det or dot and a few decimals you can check by hand. Success signal: a number comes back and matches the hand check. inv and solve produce a new matrix; still calculation, but do not start with a plot.

Current docs have no parameter tables for these functions. Do not invent extra flags. Computation runs on this machine; input matrices enter the assistant context. Do not put unpublished data in.

Use cases

Case 1: solve a linear-algebra problem. Use the listed eig, solve, svd, and so on. Fits checking eigenvalues and linear systems. Current docs only give the function-name table, no dimension limits. Symbolic work is summarized as calculus and equation solving, but no symbolic tool names are listed; if a name is missing, do not invent one.

Case 2: draw a bar chart. Call plot_barchart, described as Plots bar charts of given data values Fits turning a set of numbers into a figure. Current docs also mention plot_scatter, plot_chart, plot_stem, plot_stack, eqn_chart; on the first try plot_barchart is enough. Plotting may write a file or return an image; the path is not written. Do not assume the disk is untouched. Confirmation: bar heights match the input.

Limits

Must be Python 3.12+ and uv. The main path is a local clone plus server.py; do not treat the Smithery line npx -y @smithery/cli install @abhiphile/fermat-mcp --client gemini as the default. The author's directory is not a universal path.

Tools split into numeric, symbolic, and plotting, but symbolic tool names are almost unlist. Current docs have no auth, port, or remote deploy. Plots and matrices both enter the local session. Failure codes and image formats are not given.