Open source v2.12.0 New: meta-marketplace

Scientific tools.
In your agent.

CLIO Kit is the meta-marketplace for scientific computing. Add one MCP server, a skill or a whole workflow to Claude Code, Codex, Clio Coder or OpenCode, then check the results against your data.

Install in one line

Release
Run once
npx
Source
Agent

$uv tool install 'clio-kit>=2.11.0'

$uvx --from 'clio-kit>=2.11.0' clio-kit mcp-servers

Runs the launcher without installing it.

$npx skills@1.5.25 add iowarp/clio-kit --skill dataset-explore --agent codex --copy

Skills only, with Node.js 22.20 or newer. Use --skill '*' for every skill.

$git clone https://github.com/iowarp/clio-kit.git && cd clio-kit && uv tool install --editable .

$git clone https://github.com/iowarp/clio-kit.git && cd clio-kit

Then ask your agent: “Read setup.md and set up CLIO Kit for me.”

Need uv? curl -LsSf https://astral.sh/uv/install.sh | sh

Other ways to install →Watch the tutorials →

01 / From a question to a checked result

Ask a scientific question.
Check what comes back.

CLIO Kit gives your agent scientific tools and the procedures for using them. You choose what to connect, and every tool call stays visible for review.

01
Choose the tools
Find research, inspect data, create figures or run cluster jobs. Install the components your task needs.
02
Add the procedure
A skill describes the order of operations, scientific assumptions and checks. A plugin brings related components together.
03
Check the result
Follow the tool calls. Check source references, data coverage, job status or output files against the task’s expected result.

Clio Coder loads the dataset-explore skill and calls clio-hdf5 open_file, visit, get_shape, get_dtype, list_attributes, read_partial_dataset and close_file — A real session in Clio Coder, IOWarp’s open-source coding agent: the HDF5 tutorial’s skill checks structure and units before reading three values.

Try a first request

“Summarize these experiment results. Check units and missing values, compare the runs, and explain what the data supports.”

Choose a hands-on tutorial →

02 / Two ways to build your toolkit

Start with one component.
Bring a workflow together.

Choose a single capability or install a plugin for the task. The same launcher runs every MCP server.

One component

MCP · skill

Connect an MCP server, then copy a skill into your agent’s skill folder. Skills supply instructions; MCP servers supply tools.

$clio-kit skill install dataset-explore --target .agents/skills

Install individual components →

A workflow plugin

Bundle

A workflow plugin groups the servers and skills a task needs. One command writes the MCP configuration and copies the skills.

$clio-kit plugin install clio-scientific-io --client codex --project .

Choose a workflow →

03 / Your agent, your environment

Use the agent
you already have.

Components install into your project in each client’s own format. Every hands-on tutorial is recorded in four of them.

Connect your agent →

In the terminal: Claude Code · Codex · Clio Coder · OpenCode; Project MCP configuration, skills and visible tool calls
In your editor: Cursor · Antigravity · VS Code; Your client’s MCP configuration and skill support
Through a native marketplace: Claude Code plugins; Supported plugin dependencies, agents and hooks

Also from IOWarp

Need an agent? Clio Coder is IOWarp’s open-source coding agent for scientific software. Every tutorial here includes a Clio Coder session.

Try Clio Coder ↗

04 / Tools and knowledge for your research

A scientific
working toolkit.

Servers and skills for papers, datasets, cluster jobs, figures and I/O analysis.

Scientific data

Inspect HDF5, ADIOS and Parquet files. Read a bounded slice before working with a large dataset.

Analysis and figures

Clean tables, calculate statistics and plot results with Pandas, Plot and ParaView.

Cluster workflows

Connect to software environments, Slurm jobs and performance tools on your HPC system.

Research and discovery

Find papers, build bibliographies and discover datasets for your research.

Geoscience

Work with geographic features, terrain, seismic waveforms and event catalogues.

Performance and logs

Investigate I/O behavior, search large logs and interpret performance measurements.

Skills and agents

Add reusable procedures, scientific planning and guidance for interpreting results.

Workflow plugins

Bring related MCP servers, skills and supported agents or hooks together for a task.

Community contributions

Share your own components or index an external plugin or marketplace.

Browse all tools, skills and plugins →

05 / Install

Install CLIO Kit
and add what you need.

You need uv and an agent. Add a workflow plugin, a single MCP server or a skill; each downloads only what it needs.

Release and selective install →Client setup →

1 / Get CLIO Kit

Release
From source

uv tool install 'clio-kit>=2.11.0'

git clone https://github.com/iowarp/clio-kit.git
cd clio-kit
uv tool install --force --reinstall --editable ".[verification]"

2 / Add a component

Plugin
MCP server
Skill
Claude Code

clio-kit plugin install clio-scientific-io --client codex --project .

A workflow plugin writes its MCP configuration and copies its skills. For other agents, use --client with claude-code, opencode, cursor, antigravity or vscode.

clio-kit mcp-servers
clio-kit mcp-server hdf5

List the servers, then add the clio-kit mcp-server command to your agent’s MCP settings. The first start builds the server’s environment; some servers need system software or site access.

clio-kit skill list
clio-kit skill install dataset-explore --target .agents/skills

Use your agent’s skill folder, such as .claude/skills for Claude Code. Give --bundle clio-scientific-io instead of a name to install a workflow’s skills.

clio-kit plugin fetch clio-dataset-report --target ./clio-selected
claude plugin marketplace add "$PWD/clio-selected"
claude plugin install clio-dataset-report@clio-kit

Native Claude Code plugins can also include agents and hooks.

Find names in the catalogue.

3 / Connect your agent

Follow your agent’s setup guide, then try a tutorial with sample data.

06 / Questions

Common questions.

Do I need every server?

No. Choose individual components or a workflow. Released distributions download only the components you select. See installation options.

Can I keep using my current agent?

Yes, if it supports the component you want to use. MCP servers and portable skills work in Claude Code, Codex, Clio Coder, OpenCode and other clients. Native plugin agents and hooks depend on the host. Check your client’s setup.

What does a meta-marketplace index?

CLIO Kit’s own components and plugins from other publishers’ repositories and marketplaces, in one catalogue. Indexed code and releases stay with their authors. Meet the publishers.

Can I contribute my own tools?

Add a component or plugin here, or index a package you maintain elsewhere. Build, test and contribute a plugin.

What does a successful connection prove?

It proves the server responds. Backend setup, data access and scientific correctness still need checking. Read the validation guide.

Part of the IOWarp platform

Meet the CLIO team.

Try a workflow on data you know and inspect the result.

Built by researchers, for researchers, at Illinois Institute of Technology with NSF support.

Explore the source ↗Browse tutorials ↗Try Clio Coder ↗Part of IOWarp ↗

Try it on a dataset
you already know.

Install CLIO Kit Browse the catalogue →

Ask a scientific question.Check what comes back.

Choose the tools

Add the procedure

Check the result

Start with one component.Bring a workflow together.