SciLink MCP
An MCP server of science agents: Claude Code analyzes spectra and images, plans experiments and prepares DFT calculations
- 99 ★ on the repository
- Last commit
Broad access
Why this level
- The server executes code written by the model on the machine holding your data
- The model key is kept on the server side, and the SSE transport does not check who connects to it
Install
- Works in
Manual install
pip install scilink && scilink serve --print-mcp-json --mode both --session-dir ~/scilink_sessions/mycampaignThe command prints a secret-free MCP entry to add to `.mcp.json` or via `claude mcp add scilink -- <command>`. Put the model key in `~/.scilink/credentials.env`.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The server executes code written by the model on the machine holding your data
- The model key is kept on the server side, and the SSE transport does not check who connects to it
This is third-party code. Review the repository files before installing.
What it does
SciLink runs a research cycle for materials science and chemistry: experiment planning, data analysis and simulation. A main agent, mission control, hands the task to specialists and links their results: analysis findings feed the next experiment plan, and the new experiment's data returns for analysis. Analysis agents handle microscopy images, spectra (XPS, Raman, NMR, EELS) and hyperspectral datacubes. They write code, run it and verify the outcome, then replay the verified recipe across a whole series of files. Planning agents build hypotheses and protocols with Bayesian optimization, and simulation agents prepare structures and input files for DFT and molecular dynamics. Over MCP, SciLink works in two directions. The `scilink serve` command exposes its tools to Claude Code, Claude Desktop, VS Code Copilot and custom agents, and the `--mcp` flag connects external servers to it: instruments, databases, lab robots. A `scilink ui` web interface and a `scilink` terminal shell are available for interactive work, and every run leaves an HTML report.
Who it is for. Materials science and chemistry researchers who want to run data analysis and experiment planning from Claude Code or another agent.
Good fit when
- A spectra series must be fitted with one verified recipe to get a feature table
- Bayesian optimization suggests the next experiment point from the analysis results
- Ready input files for VASP are needed for an adsorption or defect calculation
- An approved experiment plan is turned into a protocol for a liquid-handling robot through an external MCP server
Not a fit when
- The task lies outside materials and chemistry: the specialist skills are built for them
- Model-written code must not run on the machine holding the data: the server executes it locally
- A finished simulation package with guaranteed output is needed: the simulation agents are still under development
Example request
Fit the product and byproduct bands in every UV-Vis spectrum from the data folder and suggest the next temperature and pH for maximum selectivityMCP
- Transport
- stdio, sse
- Authentication
- not required
Environment variables
ANTHROPIC_API_KEYrequired, secret- Model key, OPENAI_API_KEY or GEMINI_API_KEY also work. Kept in `~/.scilink/credentials.env` on the server side.
Limitations
A model key (Anthropic, OpenAI, Gemini or a compatible proxy) is required and kept in `~/.scilink/credentials.env` on the server side. Python 3.12 or newer is needed. The server executes code written by the model, so running it in a container or virtual machine is advised. The SSE transport does no access checks, so keep the address on `127.0.0.1` or behind an authenticating proxy. One server runs one campaign, and each campaign gets its own session folder. The README calls the simulation agents under development, and advanced segmentation needs a separate SAM install. The project was created at Pacific Northwest National Laboratory, and details are in its disclaimer.
How to disable. Run `claude mcp remove scilink` or delete the entry from `.mcp.json` and stop the `scilink serve` process. `pip uninstall scilink` removes the package, while saved sessions and memory live in the session folders and in `~/.scilink/`.
FAQ
How do I connect SciLink to Claude Code?
Run `scilink serve --print-mcp-json` to get a secret-free entry and add it to `.mcp.json` at the project root or with `claude mcp add scilink -- <command>`. The model key lives in `~/.scilink/credentials.env` on the server side.
Which modes does the server have?
The `--mode` flag picks the tool set: `analyze` for data analysis, `plan` for planning and optimization, `both` by default, `simulate` for simulation and `meta` with a single router. Take the narrowest mode for the task.
What autonomy levels does the agent have?
Three: Co-Pilot reviews every step with you, Autopilot asks you to approve major decisions, Autonomous works with no human involvement. Only the party that accepts the result changes.
What does SciLink do with external MCP servers?
It connects their tools to its orchestrators: the `--mcp` flag accepts stdio, SSE and HTTP. That is how paper search, databases, instruments and lab robots are added.
Is it safe to run the server on a work machine?
It executes code the model writes, so a container or virtual machine is safer. The SSE transport does no access checks, so keep the address on `127.0.0.1`.
Discussion
No questions or reviews yet.
Related
MCP serversMCP server that connects Claude Code and Codex to an open R or Python session in RStudio and Positron, where your data lives
An MCP server and skill for Sciverse paper search: metadata, semantic retrieval, citations and full text for agents.
SP-Mind is a Claude Agent SDK agent for spatial proteomics analysis: from image correction to cell type annotation.
A Claude Code plugin and MCP server: over 1000 scientific tools for biology, chemistry and medicine, data-check commands and routing skills
An MCP server with a toolbox for global briefs: markets, SEC filings, conflict, climate and news from open sources
A knowledge base from your documents where people and AI agents annotate together, with an MCP server and skills for Claude Code
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| SciLink (this card) | no | not required | 99 | |
| MATLAB Agentic Toolkit | no | 1,297 | a kit for agents working with MATLAB, whereas SciLink runs the analysis, planning and simulation cycle for materials and chemistry | |
| Scientific Agent Skills | no | 56,977 | a collection of skills across scientific fields, whereas here agents execute and verify the analysis code themselves |
README in short
The documentation presents SciLink as a platform for the analyze, plan and simulate cycle with three working modes and three autonomy levels. It describes foundation agents with swappable Markdown skill bundles, memory that keeps discovered techniques, knowledge bases built from papers and a Python API. A separate section explains working as both an MCP server and an MCP client.