MathKernel: An evidence-aware multi-engine mathematics kernel and MCP server
MathKernel is an open-source mathematics runtime that pairs with LLMs to handle advanced computation while preserving assumptions, provenance, and claim-specific evidence. It integrates exact, symbolic, formal, certified-interval, and numeric engines behind a typed orchestration layer, and every result carries an explicit trust level, engine tag, and derivation trail. Available as a Python library and an MCP server, it covers domains from algebra and calculus to graphs, PDEs, and algebraic topology, with tooling for visualization and sonification.
The LLM interprets intent; the MathKernel establishes mathematical evidence.
- nxobject
As a side note, the name "MathKernel" is overloaded -- that's what Mathematica's own kernel executable is called, which is a little unfortunate when you have tooling that already refers to it.
- overfitCNN
I'm building a project in this area called Alkahest: https://github.com/alkahest-cas/alkahest. It is a computer algebra system built in Rust that emits corresponding Lean certificates for many of its operations that can be formally verified.
Alkahest is available as a Python library or a Rust crate. It's designed for agent use with good debugging output.
I'm rapidly iterating on it, feel free to send any issues/comments/ideas my way.
- xvilka
You could add support also for the Woxi[1][2] for most of the symbolic math, also should be way faster than bunch of glue over different Python libraries.