IOWARP AT SC26
NOV 15–20 · CHICAGO
SC26 Get started

CLIO KIT / SCIENTIFIC TOOLS AND WORKFLOWS

Give agents real instruments.

CLIO Kit is part of CLIO, the open-source stack that IOWarp, a team at the Gnosis Research Center, builds to connect AI agents to scientific data and systems. It matters because an agent is only as useful as the tools it can reach. CLIO Kit brings together MCP servers, skills, agent definitions, hooks, and workflow plugins, packaged so a team takes what a task needs. The released launcher downloads only what you select and installs isolated server dependencies on first launch. IOWarp will launch the full CLIO stack at SC26 in Chicago.

Browse the servers Components and workflows Source
CLIO Kit · command lineMCP
  1. Install the launcheruv tool install 'clio-kit>=2.12.3'
  2. Inspect the catalogclio-kit mcp-servers
  3. Start a serverclio-kit mcp-server hdf5

The interface

MCP servers provide tools. Skills provide procedures. Workflow plugins bundle the components a task needs.

How MCP works for science
01 / The catalog

Find your tool.

  • 01 · Data

    HDF5

    Inspect datasets and attributes and compute per-index statistics.

  • 02 · Data

    Parquet

    Read columns, summarize nulls, and run grouped aggregates.

  • 03 · Data

    Pandas

    Load, filter, profile, and aggregate tabular data.

  • 04 · Data

    ADIOS

    Read slices and per-step statistics through ADIOS2 engines.

  • 05 · HPC

    Slurm

    Submit and manage cluster jobs.

  • 06 · HPC

    Spack

    Discover, install, and locate scientific packages.

  • 07 · HPC

    JARVIS

    Build and manage scientific pipelines.

  • 08 · HPC

    Darshan

    Inspect I/O performance traces.

  • 09 · HPC

    Lmod

    Manage environment modules.

  • 10 · HPC

    Parallel Sort

    Sort, filter, and summarize large log files in parallel.

  • 11 · Visual

    ParaView

    Visualize and render scientific data.

  • 12 · Visual

    Plot

    Turn tabular results into plots.

  • 13 · Earth

    Geo

    Inspect, validate, summarize, and map GeoJSON vector layers.

  • 14 · Earth

    Terrain

    Analyze elevation models and point clouds.

  • 15 · Earth

    Seismology

    Inspect SAC waveform files and archives, compute trace statistics, and analyze earthquake sequences.

  • 16 · Discovery

    ArXiv

    Find and retrieve research papers.

  • 17 · Discovery

    Scientific Catalog

    Discover operator-owned datasets.

  • 18 · Discovery

    NDP

    Search datasets across CKAN instances.

  • 19 · Discovery

    Web

    Run search and durable document fetch tasks.

  • 20 · Systems

    ChronoLog

    Log and retrieve data through ChronoLog.

  • 21 · Systems

    Compression

    Compress and decompress files.

  • 22 · Systems

    Node Hardware

    Inspect hardware on the local machine or a remote node.

Showing every server in the catalog.

02 / In practice

Choose components. Install a workflow.

Project installs support Codex, Claude Code, OpenCode, Cursor, Antigravity, and VS Code. Scientific agents and Dataset Report hooks have adapters for Codex, Claude Code, and OpenCode. Other clients can explicitly select --components-only for skills and MCP configuration; see component support.

clio-kit plugin install clio-scientific-io --client codex --project /path/to/project

Installs the scientific I/O workflow’s skills and MCP configuration into the specified Codex project. Add --dry-run to preview changes. The client setup guides explain each adapter; the selective installation guide covers release downloads and native Claude plugins.

  1. 01

    Choose

    Browse MCPs, skills, agents, and hooks, or select a workflow plugin for scientific I/O, HPC operations, analysis, or research.

  2. 02

    Install

    For an individual skill, use clio-kit skill install dataset-explore --target .agents/skills in a project whose client reads that directory. Configure its required MCP servers separately.

  3. 03

    Verify

    Reload the client and inspect its skills and tools. Run clio-kit doctor --server hdf5 --connect, then try a tool call on known data. Scientific backends and cluster access have their own prerequisites.

Next step

Make your workflow
an instrument.

Open the CLIO Kit guides CLIO Kit on the platform