MCP Numpy

What the mcp-numpy MCP server does

The mcp-numpy MCP server makes a broad subset of NumPy available as individual MCP tools. It is intended for agents that need to create arrays, transform their shapes, perform numerical calculations, or inspect array properties through an MCP client.

Array construction includes regular arrays, zero- and one-filled arrays, constant-filled arrays, ranges, evenly spaced values, identity matrices, and diagonal arrays. Manipulation tools cover reshaping, transposition, concatenation, splitting, tiling, repetition, squeezing singleton dimensions, and flattening.

The server also exposes summary statistics, minimum and maximum operations, dot and matrix multiplication, cumulative operations, differences, correlations, element-wise arithmetic, logarithms, trigonometry, and array metadata such as shape, dimensionality, size, and data type.

How it works

After installation, an MCP client starts the mcp-numpy command as a local server. The client then presents the registered NumPy operations as tools that an agent can call with operation-specific inputs. The repository documentation shows this setup for Claude Desktop and describes it as applicable to other MCP clients.

The tool set includes linear algebra operations such as matrix inversion, determinants, eigenvalues and eigenvectors, singular value decomposition, linear-system solving, and vector or matrix norms. Random-data tools generate uniform floats, normally distributed values, integers, random selections, and shuffled arrays. Statistical tools include histograms, percentiles, quantiles, correlation, and covariance-style correlation coefficients.

Setup and configuration

Install the package with pip:

pip install mcp-numpy

Add a server entry to the MCP client's configuration. The documented command-only configuration is:

{
  "mcpServers": {
    "mcp-numpy": {
      "command": "mcp-numpy"
    }
  }
}

The provided setup does not specify required environment variables, API keys, network services, or additional runtime configuration. For local development, the repository documents editable installation with test dependencies, followed by pytest, Ruff formatting and linting, and mypy checks.

Tools and capabilities

The mcp-numpy MCP server groups its tools around common NumPy workflows:

• Create arrays with np_array, np_zeros, np_ones, np_full, np_arange, np_linspace, np_eye, and np_diag.
• Change array layout with np_reshape, np_transpose, np_concatenate, np_split, np_tile, np_repeat, np_squeeze, and np_flatten.
• Run linear algebra with np_inv, np_det, np_eig, np_svd, np_solve, and np_linalg_norm.
• Generate or modify random data with np_rand, np_randn, np_randint, np_random_choice, and np_shuffle.
• Calculate statistics, including sums, means, standard deviations, variances, extrema, histograms, percentiles, quantiles, and correlations.
• Apply arithmetic, powers, modulo, roots, absolute values, exponentials, logarithms, trigonometric functions, and hyperbolic functions element by element.
• Inspect array shape, dimensions, size, and dtype, or convert types with npastype.

Limitations and notes

Choose this server for direct numerical operations; additional data-access or visualization requirements would need separate tooling.