Aman Amith Shastry Scientific Computation MCP

What Aman-Amith-Shastry-scientific_computation_mcp MCP server does

Aman-Amith-Shastry-scientific_computation_mcp MCP server gives an MCP client access to mathematical operations for vectors, matrices, tensor-like values, multivariable functions, and vector fields. Its operations cover three main areas: stored tensor manipulation, linear algebra, and vector calculus. It also returns plots for supported functions and vector fields.

The tensor store treats tensors as vectors or matrices. Clients can create named values, inspect them, and remove them. Stored matrices can be added, subtracted, multiplied, scaled, inverted, transposed, and analyzed for determinant, rank, eigenvalues, eigenvectors, QR factorization, singular value decomposition, orthonormal bases, and basis changes.

How it works

The server uses streamable HTTP and exposes its MCP endpoint at http://localhost:8081/mcp when run locally. A health endpoint is available at /health. Tensor values remain in process memory and are associated with the MCP session, allowing a sequence such as creating a tensor, viewing it, and using it in later operations.

Session state makes a single running instance important. Running multiple replicas can split the in-memory tensor store, while losing a session also loses its stored tensors. The server uses stateful HTTP sessions rather than a stateless request model.

Vector calculus tools accept mathematical expressions. Gradient, Laplacian, and directional derivative operate on scalar functions; curl, divergence, and Laplacian also support vector-field input. Vector fields must be formatted as Python-list strings for the applicable tools. Directional derivatives can normalize the supplied direction by default.

Setup and configuration

Install the server from Smithery with Node 20 or newer:

npm install -g smithery@latest
smithery mcp add @aman-amith-shastry/scientific_computation_mcp --client claude

Use --client cursor for Cursor, or omit the client option to add a remote Smithery connection. Restart the client after adding it. The Smithery namespace is lowercase and case-sensitive.

For local execution, run uv sync followed by uv run src/main.py. The server reads PORT, HOST, MCP_PATH, ALLOWED_ORIGINS, and LOG_LEVEL for runtime overrides. It does not require user configuration or a configuration schema.

Tools and capabilities

Aman-Amith-Shastry-scientific_computation_mcp includes tools for:

• Creating, viewing, and deleting named tensors.
• Performing matrix arithmetic and scaling.
• Computing inverses, transposes, determinants, ranks, eigen results, QR decompositions, SVDs, orthonormal bases, and basis changes.
• Projecting vectors and calculating dot and cross products.
• Calculating gradients, curls, divergence, Laplacians, and directional derivatives.
• Plotting supported scalar functions and three-dimensional vector fields.

Function plots support two-dimensional or three-dimensional forms based on the input variables, but only x and y can be used. Vector-field plots currently support three-dimensional fields and return a two-dimensional PNG perspective image. Plot bounds default to -1 through 1 on each axis.

Limitations and notes

The tensor store is not persistent. On hosts that sleep, tensors can disappear after the process restarts; on Render's free plan, a 15-minute period without inbound traffic can trigger this behavior. A single continuously running instance avoids the multi-replica and sleep-related state loss described in the deployment notes.

The server can be containerized and served over HTTPS by a host that supports the configured port. When publishing a URL, the health endpoint can be checked before publishing to reduce cold-start scan failures. The material does not specify a software license.