Mcp.science

🚀 MCP.science: Open Source MCP Servers for Scientific Research 🔍📚

Join us in accelerating scientific discovery with AI and open-source tools! This repository offers a collection of open-source MCP servers tailored for scientific research. These servers empower AI models to interact with scientific data, tools, and resources via a standardized protocol.

🚀 Quick Start

Running any server in this repository is as straightforward as executing a single command. For instance, to start the web-fetch server:

uvx mcp-science web-fetch

This command takes care of everything from installation to execution. For more details on configuration and finding other servers, refer to the "How to configure MCP servers for AI client apps" section below.

✨ Features

• Standardized Interaction: Enables AI models (such as Claude) to interact with scientific data, tools, and resources through a standardized protocol.
• Multiple Server Options: Offers a diverse range of MCP servers for different scientific research needs.
• Easy Configuration: Simple commands for server startup and configuration.

📦 Installation

Prerequisites

1. uv — a super-fast (Rust-powered) drop-in replacement for pip + virtualenv. Install it with:

   curl -sSf https://astral.sh/uv/install.sh | bash
   

2. An MCP-enabled client application such as Claude Desktop, VSCode, Goose, 5ire.

Starting a Server

Any server in this repository can be launched with a single shell command. The pattern is:

uvx mcp-science <server-name>

For example, to start the web-fetch stdio server locally, configure the following command in your client:

uvx mcp-science web-fetch

Which corresponds to this in claude desktop's json configuration:

{
  "mcpServers": {
    "web-fetch": {
      "command": "uvx",
      "args": [
        "mcp-science",
        "web-fetch"
      ]
    }
  }
}

The command will download the mcp-science package from PyPI and run the requested entry-point.

Find other servers

Refer to the Available servers list — every entry in the table works with the pattern shown above.

Optional: managing integrations with MCPM

MCPM is a convenience command-line tool that can automate the process of wiring servers into supported clients. It is not required but can be useful if you frequently switch between clients or maintain a large number of servers. The basic workflow is:

# Install mcpm first – it is a separate project
uv pip install mcpm

mcpm client ls           # discover supported clients
mcpm client set <name>   # pick the one you are using

# Add a server (automatically installing it if needed)
mcpm add web-fetch

After the command finishes, restart your client so that it reloads its tool configuration. You can browse the MCPM Registry for additional community-maintained servers.

💻 Usage Examples

Starting the web-fetch Server

uvx mcp-science web-fetch

📚 Documentation

Table of Contents

• About
• What is MCP?
• Available servers in this repo
• How to integrate MCP servers into LLM
• How to build your own MCP server
• Contributing
• License
• Acknowledgments
• Citation

About

This repository contains a collection of open source MCP servers specifically designed for scientific research applications. These servers enable AI models (like Claude) to interact with scientific data, tools, and resources through a standardized protocol.

What is MCP?

MCP is an open protocol that standardizes how applications provide context to LLMs. Think of MCP like a USB-C port for AI applications. Just as USB-C provides a standardized way to connect your devices to various peripherals and accessories, MCP provides a standardized way to connect AI models to different data sources and tools.

MCP helps you build agents and complex workflows on top of LLMs. LLMs frequently need to integrate with data and tools, and MCP provides:

• A growing list of pre-built integrations that your LLM can directly plug into
• The flexibility to switch between LLM providers and vendors
• Best practices for securing your data within your infrastructure

Source: https://modelcontextprotocol.io/introduction

Available servers in this repo

Below is a complete list of the MCP servers that live in this monorepo. Every entry links to the sub-directory that contains the server’s source code and README so that you can find documentation and usage instructions quickly.

Example Server

An example MCP server that demonstrates the minimal pieces required for a server implementation.

Materials Project

A specialised MCP server that enables AI assistants to search, visualise and manipulate materials-science data from the Materials Project database. A Materials Project API key is required.

Python Code Execution

Runs Python code snippets in a secure, sandboxed environment with restricted standard-library access so that assistants can carry out analysis and computation without risking your system.

SSH Exec

Allows an assistant to run pre-validated commands on remote machines over SSH with configurable authentication and command whitelists.

Web Fetch

Fetches and processes HTML, PDF and plain-text content from the Web so that the assistant can quote or summarise it.

TXYZ Search

Performs Web, academic and “best effort” searches via the TXYZ API. A TXYZ API key is required.

Timer

A minimal countdown timer that streams progress updates to demonstrate MCP notifications.

GPAW Computation

Provides density-functional-theory (DFT) calculations through the GPAW package.

Jupyter-Act

Lets an assistant interact with a running Jupyter kernel, executing notebook cells programmatically.

Mathematica-Check

Evaluates small snippets of Wolfram Language code through a headless Mathematica instance.

NEMAD

Neuroscience Model Analysis Dashboard server that exposes tools for inspecting NEMAD data-sets.

TinyDB

Provides CRUD access to a lightweight JSON database backed by TinyDB so that an assistant can store and retrieve small pieces of structured data.

How to configure MCP servers for AI client apps

If you're not familiar with these stuff, here is a step-by-step guide for you: Step-by-step guide to configure MCP servers for AI client apps

How to build your own MCP server

Please check How to build your own MCP server step by step for more details.

🔧 Technical Details

Server Startup

The uvx command is used to start the servers. It downloads the mcp-science package from PyPI and runs the requested entry-point.

Configuration

Servers can be configured using shell commands or JSON configuration files in client applications.

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

Thanks to all contributors!

Citation

For general use, please cite this repository as described in the root CITATION.cff. If you use a specific server/subproject, please see the corresponding CITATION.cff file in that subproject's folder under servers/ for the appropriate citation.

Contributing

We enthusiastically welcome contributions to MCP.science! You can help with improving the existing servers, adding new servers, or anything that you think will make this project better. If you are not familiar with GitHub and how to contribute to an open source repository, it might be a bit challenging, but it's still easy for you. We would recommend you to read these first:

• How to make your first pull request on GitHub
• Creating a pull request In short, you can follow these steps:

1. Fork the repository to your own GitHub account
2. Clone the forked repository to your local machine
3. Create a feature branch (git checkout -b feature/amazing-feature)
4. Make your changes and commit them (git commit -m 'Add amazing feature')
   👈 Click to see more conventions about directory and naming

   Please create your new server in the `servers` folder. For creating a new server folder under the repository folder, you can simply run (replace `your-new-server` with your server name) ```sh uv init --package --no-workspace servers/your-new-server uv add --directory servers/your-new-server mcp ``` This will create a new server folder with the necessary files: ```bash servers/your-new-server/ ├── README.md ├── pyproject.toml └── src └── your_new_server └── __init__.py ``` You may find there are 2 related names you might see in the config files: 1. **Project name** (hyphenated): The folder, project name and script name in `pyproject.toml`, e.g. `your-new-server`. 2. **Python package name** (snake_case): The folder inside `src/`, e.g. `your_new_server`.
5. Push to the branch (git push origin feature/amazing-feature)
6. Open a Pull Request Please make sure your PR adheres to:

• Clear commit messages
• Proper documentation updates
• Test coverage for new features

Contributor Recognition in Subrepos

If you want to recognize contributors for a specific server/subrepo (e.g. servers/gpaw-computation/), you can use the All Contributors CLI in that subdirectory. Steps:

1. In your subrepo (e.g. servers/gpaw-computation/), create a .all-contributorsrc file (see example).
2. Add contributors using the CLI:

   npx all-contributors add <github-username> <contribution-type>
   

3. Generate or update the contributors section in the subrepo's README.md:

   npx all-contributors generate
   

4. Commit the changes to the subrepo's README.md and .all-contributorsrc. For more details, see the All Contributors CLI installation guide.