HDF5

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

Install the launcher first. Restart your client after setup. Configuration locations →

clio-kit doctor --server hdf5 --connect

Tools​

open_file

Open an HDF5 file for operations.

close_file

Close the current HDF5 file.

Details

Returns: Status message

get_filename

Get the current file's path.

Details

Returns: File path

get_mode

Get the current file's access mode.

Details

Returns: File mode

get_by_path

Get a dataset or group by path.

list_keys

List keys in a group.

visit

Visit all nodes recursively.

read_full_dataset

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

read_partial_dataset

Read a portion of a dataset with slicing.

get_shape

Get the shape of a dataset.

get_dtype

Get the data type of a dataset.

get_size

Get the size of a dataset.

get_chunks

Get chunk information for a dataset.

read_attribute

Read an attribute from an object.

list_attributes

List all attributes of an object.

hdf5_parallel_scan

Fast multi-file scanning with parallel processing.

hdf5_batch_read

Read multiple datasets in parallel.

hdf5_stream_data

Stream large datasets efficiently with memory management.

hdf5_aggregate_stats

Parallel statistics, with explicit sampling coverage above 500 MiB.

Details

Sample statistics describe only the selected values, not full-dataset totals.

analyze_dataset_structure

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

find_similar_datasets

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

suggest_next_exploration

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

identify_io_bottlenecks

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

optimize_access_pattern

Suggest better approaches for data access based on usage patterns.

refresh_hdf5_resources

Re-scan configured discovery directories; explicit paths can be opened separately.

Details

FastMCP automatically sends notifications/resources/list_changed to clients. Returns: Summary of refreshed resources

list_available_hdf5_files

List discovered HDF5 files; open_file(path) also accepts unlisted explicit paths.

Details

Returns: List of available files with resource URIs

export_dataset

Export dataset to various formats with user format selection.

Examples​

Replace example paths with your own files and check the returned results.

Inspect a file before reading data​

Call open_file, then list_keys, get_shape and get_dtype for the selected dataset. Read a bounded preview with read_partial_dataset; start and count are comma-separated strings, such as "0,0" and "5,4". Use export_dataset with an explicit export_format only when the dataset fits in memory, and call close_file after the workflow.

The runnable HDF5 example creates a known dataset, calls the installed server and verifies exported values.

Current state and coverage limits​

Open the file before hdf5_batch_read or hdf5_aggregate_stats; their paths refer to datasets inside the current file. Above 500 MiB, aggregate statistics may use a strided sample. Responses label SAMPLED versus FULL DATA, give processed/total element counts and sampled coverage, and warn that sample totals and bounds do not describe the whole dataset. Cross-dataset aggregation is omitted if any result is sampled. Streaming stops at max_chunks, and multidimensional chunks slice the first axis. Check memory requirements, sampling and processed coverage before treating the result as an exact whole-dataset statistic.