MCP.science

MCP.science is an MCP server suite that connects large language models and AI clients to specialized scientific tools, computation engines, and research databases. Developed by the Path Integral Institute, this open-source monorepo provides researchers, data scientists, and computational academics with standardized interfaces to conduct scientific analysis directly through AI assistants like Claude. The collection includes modular servers targeting distinct scientific workflows. It connects assistants to domain-specific platforms such as the Materials Project database for materials-science calculations and GPAW for density-functional-theory computations. It also provides tools for running sandboxed Python code snippets, interacting programmatically with active Jupyter notebook kernels, and evaluating Wolfram Language code through headless Mathematica instances. In addition, MCP.science supplies utility servers for fetching web documents and PDFs, performing academic searches via the TXYZ API, executing remote commands over SSH, and storing lightweight structured data with TinyDB. By providing a unified protocol for scientific tooling, it enables researchers to orchestrate automated computational experiments, query academic knowledge bases, and manipulate datasets without manual context switching between separate terminal utilities and analytical environments.

Category: Developer Tools & Code Intelligence

Tags: academic, data analysis, research, science

Visit MCP.science

How to install and configure MCP.science

1. Install the uv tool on your system: bash curl -sSf https://astral.sh/uv/install.sh | bash 2. Identify the server module you want to run, such as web-fetch, python-code-execution, or materials-project. 3. Open your MCP client configuration file (for example, Claude Desktop's JSON settings). 4. Register the server under mcpServers using uvx: json { "mcpServers": { "web-fetch": { "command": "uvx", "args": [ "mcp-science", "web-fetch" ] } } } 5. Restart your MCP client application to load the server and begin using its tools.

What you can do with MCP.science

• Query and retrieve materials-science datasets, properties, and visual structures directly from the Materials Project database. - Run density-functional-theory quantum chemistry computations programmatically within AI assistant workflows via the GPAW computation server. - Connect language models to active Jupyter kernels to evaluate notebook cells and inspect analytical output. - Submit symbolic equations and math routines to a headless Mathematica engine for automated verification. - Search research literature and retrieve academic full texts using the TXYZ search integration.

Key facts

• Open Source
• https://github.com/pathintegral-institute/mcp.science
• Developer Tools & Code Intelligence, Web Search & Research
• academic, data analysis, research, science

Part of MCP Servers

How do I install MCP.science?

You can install and run any server from MCP.science using the uv package runner. First install uv on your system, then launch your selected server by running uvx mcp-science followed by the server name, such as web-fetch or python-code-execution. In client applications like Claude Desktop, you configure the command as uvx with arguments specifying mcp-science and the server identifier.

What can MCP.science do?

MCP.science provides specialized MCP servers covering multiple areas of research and data analysis. It enables AI assistants to search the Materials Project database, execute Python snippets in a sandbox, run DFT calculations through GPAW, interact with live Jupyter kernels, evaluate Wolfram Language code via Mathematica, fetch online research papers, and manage local TinyDB databases.

Which MCP clients work with MCP.science?

MCP.science works with any client application that implements the Model Context Protocol over standard input and output. Tested and supported clients include Claude Desktop, Visual Studio Code with MCP extensions, Goose, and 5ire. You can also automate server configuration and management across supported clients using the third-party MCPM command-line utility.

Is MCP.science open source?

Yes, MCP.science is open-source software released under the permissive MIT license. The source code, individual server modules, and step-by-step guides are publicly maintained on GitHub by the Path Integral Institute, welcoming community contributions for new scientific integrations and server improvements.