CLIO Kit

Comprehensive toolkit for building autonomous AI agents with CLIO Kit. 15+ MCP servers, 150+ tools, evaluation frameworks, observability platform.

MCP Servers

Specialized servers for scientific computing, HPC, and data management. Each server provides domain-specific tools accessible via Model Context Protocol.

CLIO Kit dashboard showing MCP server grid: HDF5, Slurm, Darshan, ParaView, and more

mcp-hdf5

Read, write, and query HDF5 scientific data files. Access datasets, attributes, and groups.

mcp-slurm

Submit, monitor, cancel HPC jobs. Query queue status and node availability.

mcp-darshan

Analyze I/O profiling data from Darshan logs.

mcp-lmod

Load/unload environment modules on HPC systems.

mcp-jarvis

Deploy and manage complex software pipelines with Jarvis framework.

mcp-node-hardware

Query node hardware: CPUs, GPUs, memory, storage topology.

mcp-pandas

Analyze tabular data with pandas DataFrame operations.

mcp-parquet

Read/write Apache Parquet columnar data files.

mcp-arxiv

Search and retrieve scientific papers from arXiv.

mcp-compression

Compress/decompress data with scientific-optimized codecs.

mcp-parallel

Execute parallel computations across distributed nodes.

mcp-plot

Generate publication-quality plots and visualizations.

mcp-adios

Interface with ADIOS2 streaming I/O framework.

mcp-chronolog

Distributed shared log for multi-agent coordination and event ordering.

mcp-ndp

Near-Data Processing for in-situ computation at storage layer.

mcp-paraview

ParaView scientific visualization and rendering.

Installation

Install MCP servers for Claude Code, Cursor, or VS Code. Choose your preferred IDE below.

# Claude Code
claude mcp add clio-hdf5 -- uvx clio-kit mcp-server hdf5

# Cursor (.cursor/mcp.json)
{
  "mcpServers": {
    "clio-hdf5": {
      "command": "uvx",
      "args": ["clio-kit", "mcp-server", "hdf5"]
    }
  }
}

# VS Code (settings.json)
"mcp.servers": {
  "clio-hdf5": {
    "command": "uvx",
    "args": ["clio-kit", "mcp-server", "hdf5"]
  }
}

Meta-MCP ProtocolPlanned

IOWarp's Meta-MCP server dynamically routes requests to the appropriate specialized server based on data format and task requirements. One connection, all capabilities.

Instead of managing multiple MCP server connections, connect once to the Meta-MCP server. It intelligently delegates to mcp-hdf5 for HDF5 files, mcp-slurm for job scheduling, mcp-darshan for I/O profiling, and so on. This unified interface simplifies agent development while maintaining full access to all 150+ specialized tools.

CLIO EvalIn Development

Test agent capabilities across scientific domains. Measure context quality, tool accuracy, reasoning chains.

The evaluation framework provides comprehensive benchmarks for assessing agent performance in scientific computing scenarios. Evaluate how well agents understand context, execute tool calls correctly, maintain reasoning coherence, and adapt to domain-specific challenges. Built-in metrics track success rates, latency, resource utilization, and context retention across multi-step workflows.

▸ Context Quality: Measure how well agents preserve and utilize context across interactions
▸ Tool Accuracy: Validate correctness of tool invocations and data transformations
▸ Reasoning Chains: Analyze multi-step problem-solving workflows and decision paths
▸ Domain Coverage: Test across HPC, climate science, materials research, and genomics

Observability PlatformPlanned

OpenTelemetry-based tracing, real-time dashboards, agent behavior monitoring.

Gain deep insights into agent operations with production-grade observability. Track every tool call, context transfer, and decision point. Real-time dashboards visualize agent workflows, performance metrics, and resource utilization. Built on OpenTelemetry standards for seamless integration with existing monitoring infrastructure.

Distributed Tracing

End-to-end request tracing across agents, tools, and storage tiers. Identify bottlenecks and optimize workflows.

Real-Time Metrics

Live performance dashboards showing latency, throughput, error rates, and context utilization.

IDE Integration

Works seamlessly with Claude Code, Cursor, and VS Code. Install MCP servers once, use them across all supported platforms.

Anthropic

Claude Code

Cursor

VS Code