Math MCP Server

Math MCP Server is a Python 3.12 Model Context Protocol server that exposes a broad mathematics toolkit over stdio. It combines symbolic algebra, calculus, numerical methods, linear algebra, statistics, discrete mathematics, graph theory, rendering, and optional GPU acceleration behind MCP tools that are usable from Claude Desktop, GitHub Copilot, and other MCP-compatible hosts.

The server is designed to run on CPU-only systems by default. If CUDA and CuPy are available, the GPU tool family accelerates supported workloads automatically and otherwise falls back to CPU implementations when possible.

Installation

Editable install with uv:

uv pip install -e .

For GPU acceleration (requires CUDA 12.x and a compatible NVIDIA GPU):

uv pip install -e ".[gpu]"

Editable install with uv and optional GPU support:

uv pip install -e ".[gpu]"

Editable install with pip:

pip install -e .

Use the base install for CPU-only systems. Install the gpu extra only on machines with a compatible CUDA 12.x environment.

For development dependencies and tests:

.venv/bin/uv sync --extra dev

Running The Server

Run directly with Python:

python -m math_mcp.server

Run through uv:

uv run python -m math_mcp.server

Run via the console script:

math-mcp-server

The server communicates over stdio, which is the expected transport for most MCP hosts.

MCP Host Configuration

Most MCP clients accept a JSON config that declares a named server, the command to run, and its arguments. A generic config looks like this:

{
	"mcpServers": {
		"math": {
			"command": "/absolute/path/to/.venv/bin/python",
			"args": ["-m", "math_mcp.server"],
			"cwd": "/absolute/path/to/math-mcp-server"
		}
	}
}

If the package is already installed in the active environment, you can use the console script instead:

{
	"mcpServers": {
		"math": {
			"command": "math-mcp-server"
		}
	}
}

Claude Desktop

Add the math server entry to your Claude Desktop MCP configuration file and point it at either the environment Python executable or the installed math-mcp-server script.

GitHub Copilot

Use the same mcpServers structure in your MCP-enabled Copilot or VS Code host configuration. The server name can remain math so it matches the packaged MCP entry point.

Other MCP Hosts

Any host that supports stdio MCP servers can run this project with the same command and args shown above.

Tool Catalog

Symbolic Tools

Tool	Description
simplify	Simplify algebraic, trigonometric, and rational expressions.
expand	Expand products, powers, and composite symbolic expressions.
factor	Factor polynomials and integers into structured symbolic factors.
solve	Solve equations, systems, and inequalities over configurable domains.
symbolic_integrate	Compute indefinite and definite symbolic integrals.
symbolic_diff	Differentiate expressions, including higher-order derivatives.
limit	Evaluate one-sided and two-sided limits.
series	Produce Taylor or Laurent-like series expansions.
symbolic_sum	Evaluate finite and infinite symbolic summations.
symbolic_product	Evaluate finite and infinite symbolic products.
dsolve	Solve ordinary differential equations symbolically.
laplace_transform	Compute Laplace transforms in the frequency domain.
inverse_laplace	Transform Laplace-domain expressions back to time-domain forms.

Numerical Tools

Tool	Description
numerical_integrate	Numerically integrate real-valued functions on finite or improper intervals.
find_root	Find roots with bracketing or open methods such as Brent or Newton.
ode_solve	Solve initial value problems for scalar or vector ODE systems.
interpolate_data	Interpolate tabular data with linear or spline-based methods.
curve_fit	Fit parametric models to observed data and report fitted parameters.

Linear Algebra Tools

Tool	Description
matrix_multiply	Multiply dense matrices and vectors.
solve_linear	Solve square linear systems with direct or iterative methods.
lstsq	Solve overdetermined systems with least-squares regression.
matrix_decompose	Compute QR, LU, Cholesky, or SVD-style decompositions.
eigen_decomp	Compute eigenvalues and eigenvectors.
matrix_info	Report determinant, rank, conditioning, definiteness, and shape metadata.

Statistics Tools

Tool	Description
describe	Return descriptive statistics for a sample.
distribution	Evaluate distribution PDFs, CDFs, quantiles, and related functions.
hypothesis_test	Run supported hypothesis tests and report significance details.
regression	Fit linear and polynomial regression models.
bootstrap	Estimate confidence intervals with bootstrap resampling.
random_sample	Draw samples from supported probability distributions.

Discrete Math And Number Theory Tools

Tool	Description
factor_int	Factor integers into prime powers.
is_prime	Test whether an integer is prime.
primes_range	Enumerate primes over an interval or generate the first n primes.
gcd_lcm	Compute GCD, LCM, extended GCD, or Euclidean algorithm steps.
modular	Solve modular arithmetic tasks such as inverses, powers, and congruences.
combinatorics	Evaluate permutations, combinations, Catalan numbers, and related counts.
partitions	Count integer partitions and constrained partition variants.
diophantine	Solve supported Diophantine equations over the integers.

Graph Theory Tools

Tool	Description
graph_create	Build graphs from named templates, edge lists, or adjacency matrices.
shortest_path	Compute shortest paths on weighted or unweighted graphs.
spanning_tree	Compute a spanning tree or minimum spanning tree.
graph_metrics	Report graph connectivity and structural metrics.
max_flow	Compute maximum flow on directed graphs.

GPU Tools

Tool	Description
gpu_matrix_multiply	Accelerate matrix multiplication with GPU support and CPU fallback.
gpu_fft	Compute FFTs on GPU when available.
gpu_eigen_batch	Compute batched eigenvalues for multiple matrices.
gpu_solve	Solve batched linear systems using GPU acceleration where available.

Rendering Tools

Tool	Description
to_latex	Convert symbolic expressions to LaTeX and renderable output.
render_math	Render LaTeX math to an image data URI.
plot_function	Plot explicit functions over configurable domains.
plot_implicit	Plot implicit equations and inequalities.

Hardware Requirements

• Python 3.12 or newer.
• CPU-only systems are fully supported for the non-GPU tool families.
• Optional CUDA-capable NVIDIA GPU for the GPU tool family.
• If CUDA or CuPy is unavailable, GPU tools either fall back to CPU implementations or report that GPU acceleration is not available.
• Memory needs depend on workload size. Large symbolic expressions, large matrices, and graph algorithms can require significant RAM.

Example Queries

• Simplify (x^2 - 1) / (x - 1).
• Solve x^2 - 5x + 6 = 0 over the reals.
• Compute the definite integral of sin(x) from 0 to pi.
• Find the eigenvalues of [[4, 1], [1, 3]].
• Fit an exponential decay model to observed time-series data.
• Compute the bootstrap confidence interval for a sample mean.
• Factor 123456789 into prime powers.
• Find the shortest weighted path from node A to node D.
• Render the equation x^2 + y^2 = 1 as an implicit plot.
• Multiply a batch of matrices with GPU acceleration if hardware is available.

Verification

uv run python -c "from math_mcp.server import main; print('Import OK')"
uv run python -c "from math_mcp.tools.symbolic import simplify; print('Tools OK')"
uv run pytest tests/ -v