mathguard
Check LLM maths with SymPy and Pint: identities, derivations, calculus and units
- Platforms
- CLI, API
- Runs
- Locally
- Version
- 0.4.0
- Listing updated
- today
Similar MCP servers
About mathguard
mathguard is an MCP server for education and exams. It runs locally on your machine and is started by your AI client.
Language models set up calculations well but often slip when carrying them out. mathguard is a verification layer built on SymPy and Pint that checks a claim and, when it is wrong, gives a concrete counterexample. It can test a single identity, walk a derivation step by step and point at the faulty line, or grade a model's answer against a reference with unit conversion and rounding tolerance.
Results use deliberately honest statuses. Verified means a computer algebra system confirmed the claim; likely means many random points agreed without a proof; refuted, inconclusive and error cover the rest. The README notes that verified is not formal proof and points to Lean 4 for that.
A whole worked solution written in Markdown or LaTeX can be checked in one go, and the server also exists as a Python library and command-line tool (mathguard doc, mathguard grade). A browser demo is hosted on Hugging Face. It installs from PyPI.
Install
Run in a terminal
claude mcp add mathguard -- uvx mathguardAdd to claude_desktop_config.json
{
"mcpServers": {
"mathguard": {
"command": "uvx",
"args": [
"mathguard"
]
}
}
}Add to ~/.cursor/mcp.json
{
"mcpServers": {
"mathguard": {
"command": "uvx",
"args": [
"mathguard"
]
}
}
}Add to .vscode/mcp.json
{
"servers": {
"mathguard": {
"type": "stdio",
"command": "uvx",
"args": [
"mathguard"
]
}
}
}Add to ~/.codeium/windsurf/mcp_config.json
{
"mcpServers": {
"mathguard": {
"command": "uvx",
"args": [
"mathguard"
]
}
}
}Add to your client's MCP settings (most use this shape)
{
"mcpServers": {
"mathguard": {
"command": "uvx",
"args": [
"mathguard"
]
}
}
}Using several servers? Build one config file with the MCP config generator.
Tools this server provides (11)
Taken from the project documentation; the running server may differ.
- check_equal
- Tests whether two expressions are equal.
- check_chain
- Checks each step of a derivation.
- check_derivative
- Differentiates and compares.
- check_antiderivative
- Differentiates a claimed antiderivative.
- check_definite_integral
- Verifies a definite integral.
- check_solutions
- Checks claimed solutions and missing ones.
- check_value
- Evaluates a numeric claim to tolerance.
- check_limit
- Checks a limit.
- check_answer
- Grades final results against a reference.
- check_document
- Checks a whole Markdown or LaTeX solution.
- check_dimensions
- Dimensional analysis with units.
What it does
- Step-by-step derivation check
- Finds the first wrong step in a chain of rewritings.
- Answer grading
- Compares formulas symbolically and numbers with unit conversion.
- Calculus checks
- Verifies derivatives, antiderivatives, definite integrals and limits.
- Solution set check
- Substitutes claims and flags missing solutions.
- Dimensional analysis
- Detects unit mismatches with Pint.
- Whole-document check
- Reads LaTeX or Markdown solutions and reports per step.
Good for
- Verify a model's worked physics solution
- Grade LLM answers in an evaluation against reference results
- Confirm an integral or derivative claim
- Catch a missing root in a solved equation
- Detect inconsistent units in an equation
mathguard FAQ
Does verified mean proven?
No. It means a computer algebra system confirmed it; the README points to Lean 4 for machine-checked proofs.
How do I install it?
pip install mathguard.
What formats can it read?
Plain text, Markdown and LaTeX, including align and equation environments.
Can I try it without installing?
Yes, there is a Hugging Face demo.
How do I add mathguard to Claude Code?
Run this command in your terminal: claude mcp add mathguard -- uvx mathguard
What are the best alternatives to mathguard?
The closest alternatives to mathguard are REA, Context7, Chrome DevTools MCP, Codebase Memory and GitHub.
Where this comes from
- Description: written from the project's README.
- Install steps: from the official MCP Registry entry.


