HDF5 MCP Server

Community

HDF5 scientific data access for AI agents with AI-powered insights.

How to install ↓ Compare GitHub source repository ↗

Category: Filesystem #6 of 24
Stars: ★ 30 Popular
Transport: stdio (local process)
Runtime: Python 3.10+ · Prebuilt binary
Credentials: No credential needed
License: BSD-3-Clause
Last commit: 2 months ago
Tools: 27

63FMRS · C

This server provides comprehensive HDF5 data access tools for AI agents, well-suited for scientific computing scenarios. Verified that the tool descriptions align with the actual functionality; official status remains to be confirmed.

Strongest · Setup experience 14/20 Weakest · Reliability 10/20

Reliability

10/20

Security and permissions

13/20

Maintenance

14/20

Documentation

12/20

Setup experience

14/20

Why each score

Reliability 10/20: Static review cannot verify the hdf5 server starts and completes the MCP handshake; the manifest declares 27 tools, 3 resource templates and 4 prompts with coherent descriptions, but no server-specific tests or execution evidence are committed — only repo-level CI/coverage badges. Version inconsistencies (server 2.2.3, package clio-kit 2.10.6, README install example 2.4.3) further reduce confidence. Deducted for unverifiable key paths and version drift.
Security and permissions 13/20: No red-line risk: tools are predominantly read/analyze (open_file, read_*, list_*, stats); no malware, credential theft, or irreversible delete/deploy defaults found. However, open_file/export_dataset accept arbitrary file paths with no disclosed sandboxing, root constraint, or confirmation for write-side effects (export, parallel scan output). Deducted for missing least-privilege scoping and data-boundary disclosure.
Maintenance 14/20: Active, non-archived repo with 39 open issues, badges for CI, ruff, mypy and pip-audit, clear BSD-3-Clause license, identified owner (GRC at IIT) with NSF sponsorship and community channels. Deducted for version governance gaps (manifest vs package vs README versions disagree) and lack of server-level dependency-update evidence.
Documentation 12/20: README is well layered: install steps, four client config examples, troubleshooting, HDF5 usage examples, and lockfile mechanism notes. But the 27 tools have only brief descriptions — no parameter schemas, return structures, or disclosed limits (large-dataset preview truncation, export formats). Manifest/README version mismatch can mislead users. Deducted for tool-level parameter docs and limitation disclosure.
Setup experience 14/20: Setup path is clear: uv tool install clio-kit then 'clio-kit mcp-server hdf5' over stdio, with complete JSON configs for Cursor, Claude Code, VS Code and Claude Desktop, and platform notes (Python 3.10+, Linux/macOS/Windows uv). Deducted because packageArguments in the manifest must be manually reconciled with README examples, and no execution evidence verifies out-of-the-box success; static calibration caps this at 15.

Static review · not runListed 2026-09-08

Read the FMRS scoring method →

Fit and risk

What it can accessReads local filesWrites / deletes local files

Best for

Scientists
Data scientists
AI agents working on scientific computing

Not for

Scenarios where HDF5 files are not used
Users who want to directly manipulate databases or CSV files

Required permissions

Access to local file system to read HDF5 files
May create temporary files or caches
Network access if remote files are required

Risks and side effects

Potential memory issues with very large HDF5 files
Parallel operations may cause I/O contention

Setup

Before you start

Runtime：Python 3.10+ · Prebuilt binary

Install the clio-kit package using pip or uv tool, then run 'clio-kit mcp-server hdf5'.

claude_desktop_config.json

{
  "mcpServers": {
    "hdf5-mcp": {
      "command": "clio-kit",
      "args": ["mcp-server", "hdf5"]
    }
  }
}

Shown for Claude Desktop. Other clients may use a different file or key (VS Code uses "servers") — the configurator below converts it.

.vscode/mcp.json

{
  "servers": {
    "hdf5-mcp": {
      "command": "clio-kit",
      "args": [
        "mcp-server",
        "hdf5"
      ]
    }
  }
}

Goes in your project's .vscode/mcp.json (VS Code uses a "servers" key).

Terminal

claude mcp add hdf5-mcp -- clio-kit mcp-server hdf5

Run it in a terminal; replace any <…> placeholders with your own values first.

Generate and validate client config

Check that it works

Run `clio-kit mcp-server hdf5` and connect your client with the hdf5-mcp config; confirm tools like open_file and list_keys appear in the client's tool list, then ask "What datasets are in this HDF5 file?" to prove the connection.

Troubleshooting

Verify the server command is correct, ensure the HDF5 file exists and is readable, and check logs for I/O errors.

Things to try

Once connected, you can ask your AI assistant things like:

Open climate_simulation.h5, analyze the file structure, and list all datasets.
Read the first 100 timesteps of the temperature dataset.
Compute parallel aggregate statistics across the pressure datasets and identify any I/O bottlenecks.
Export the current dataset to CSV.

Tools 27

open_file read-only

Open an HDF5 file for operations.

close_file read-only

Close the current HDF5 file.

get_filename read-only

Get the current file's path.

get_mode read-only

Get the current file's access mode.

get_by_path read-only

Get a dataset or group by path.

list_keys read-only

List keys in a group.

visit read-only

Visit all nodes recursively.

read_full_dataset read-only

Read an entire dataset with efficient chunked reading for large datasets.

Show 19 more tools

read_partial_dataset read-only

Read a portion of a dataset with slicing.

get_shape read-only

Get the shape of a dataset.

get_dtype read-only

Get the data type of a dataset.

get_size read-only

Get the size of a dataset.

get_chunks read-only

Get chunk information for a dataset.

read_attribute read-only

Read an attribute from an object.

list_attributes read-only

List all attributes of an object.

hdf5_parallel_scan read-only

Fast multi-file scanning with parallel processing.

hdf5_batch_read read-only

Read multiple datasets in parallel.

hdf5_stream_data read-only

Stream large datasets efficiently with memory management.

hdf5_aggregate_stats read-only

Parallel statistics computation across multiple datasets.

analyze_dataset_structure read-only

Analyze and understand file organization and data patterns with AI insights.

find_similar_datasets read-only

Find datasets with similar characteristics to a reference dataset with AI analysis.

suggest_next_exploration read-only

Suggest interesting data to explore next based on current location with AI recommendations.

identify_io_bottlenecks read-only

Identify potential I/O bottlenecks and performance issues with AI recommendations.

optimize_access_pattern read-only

Suggest better approaches for data access based on usage patterns.

refresh_hdf5_resources read-only

Re-scan client roots and update available HDF5 resources.

list_available_hdf5_files read-only

List all registered HDF5 files with resource URIs for Claude Code @ mentions.

export_dataset writes

Export dataset to various formats with user format selection.

Use cases

AI agents analyzing scientific data in HDF5 files such as temperature fields.

Automated data exploration and structural analysis.

Parallel reading of multiple HDF5 datasets to accelerate workflows.

AI-assisted optimization of data access patterns.

Supported clients

Claude Desktop

Listed from the project's documentation, not tested by this site.

Overview

The HDF5 MCP server is a Model Context Protocol server for scientific data access, part of CLIO Kit. It provides a comprehensive set of tools that enable AI agents to open, read, query, and analyze HDF5 files, featuring efficient chunked reading, parallel operations, and AI-assisted data structure analysis and optimization suggestions.

Similar servers

Filesystem MCP Server 84 · B

Secure local file read/write access

★ 91.4k · Tools 6 Compare with this →

Vault Cortex MCP Server 79 · B

Standalone MCP server for Obsidian vaults: hybrid search, notes & files, structured memory, tasks, OAuth 2.1.

★ 30 · Tools 30 Compare with this →

Obsidian Web MCP 68 · C

Secure remote access to your Obsidian vault over HTTPS with OAuth 2.0 authentication.

★ 185 · Tools 20 Compare with this →

Safe Docx MCP Server 68 · C

AI-native surgical editing of Word and OpenDocument files with formatting preservation

★ 46 Compare with this →

Compare with other MCP servers

More servers from this repository

Source revision 93e41d70d6c2 Data synced 2026-10-19 Read the FMRS scoring method