M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
C

CLIO Parquet

Community
MCP server for Apache Parquet files
How to install ↓ Compare GitHub source repository ↗
Category
Filesystem #14 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
Tools
4
55FMRS · C

CLIO Parquet is a specialized MCP server for reading Apache Parquet files. It provides essential tools for summarizing, row reading, column preview, and aggregation statistics. Being part of the CLIO Kit ecosystem, it supports scientific and HPC contexts. Its functionality is clear but limited; it is suitable for its targeted use of summarization and reading, but may not cover complex full data analysis needs.

Strongest · Maintenance 13/20 Weakest · Reliability 8/20

Reliability
8/20
Security and permissions
12/20
Maintenance
13/20
Documentation
10/20
Setup experience
12/20
Why each score
Reliability 8/20
The manifest's four tools (summarize/read_slice/column_preview/aggregate) are coherent with Parquet semantics and the package arguments (mcp-server parquet) are plausible. However, no committed tests, real tools/list transcripts, or execution evidence specific to the parquet server is supplied; the README's machine-readable contracts cover only JARVIS/SLURM/Spack/Scientific Catalog, not parquet. The manifest version 2.2.3 also mismatches the PyPI identifier version 2.10.6, weakening version traceability. Under static review this cannot exceed the calibration cap.
Security and permissions 12/20
All tools are read-only local Parquet analysis (schema, row slices, column preview, aggregation) over stdio transport, with no network calls, credential handling, or destructive operations; no red-line behaviors observed. But no source is provided to verify path validation, error handling, or data boundaries, and permission scoping/data-flow disclosure is incomplete, so higher marks are not justified.
Maintenance 13/20
The repository is not archived, carries a clear BSD-3-Clause license, and shows CI, coverage, and pip-audit badges, NSF sponsorship, named institutional ownership (GRC/IIT, HDF Group, Utah), and active community channels (Zulip, Issues, Discussions) plus live PyPI releases. However, versions contradict each other across manifest (2.2.3), README install text (2.4.3), and package (2.10.6), and per-server maintenance cadence for parquet cannot be confirmed, so governance/versioning gaps are deducted.
Documentation 10/20
The README gives overall installation, multi-client config examples, and a troubleshooting section, and parquet has a one-row entry in the server table. However, none of the four parquet tools has documented parameters, return shapes, limits, or filter syntax, and no server-specific README or usage example exists — usable but with many hidden assumptions.
Setup experience 12/20
The install path is clear: uv tool install clio-kit then clio-kit mcp-server parquet, with manifest registryType/identifier and positional package arguments directly usable by clients. Yet the README's client JSON examples omit parquet, the wheel launches via an embedded uv.lock mechanism that adds fragility, and documented versions are inconsistent, so setup stays below the static ceiling.

Static review · not runListed 2026-09-08

Read the FMRS scoring method →

Fit and risk

What it can accessReads local files

Best for

  • Data exploration in scientific computing and HPC environments.
  • AI agents that need to quickly summarize Parquet files.
  • Workflows integrated with the CLIO Kit ecosystem, where other servers provide complementary functions.

Not for

  • Full-fledged data analysis with complex transformations (e.g., group by aggregates), which might require dedicated tools.
  • Handling extremely large files where the entire data cannot fit into memory for operations like reading all rows.

Required permissions

  • Read access to Parquet files in the local file system as provided by the user.
  • Perform file I/O to read file metadata and contents.

Risks and side effects

  • Potential exposure of sensitive data if inputs include unexpected file paths.
  • Resource-intensive operations on large files or complex queries could lead to excessive memory or CPU consumption.

Setup

Before you start

Runtime：Python 3.10+ · Prebuilt binary

  1. Ensure you have Python 3.10+ and the uv package manager installed.\n2. Install clio-kit with uv tool install 'clio-kit==2.10.6'.\n3. Run uv tool update-shell if the executable directory is not on PATH.\n4. Verify server list: clio-kit mcp-servers.\n5. Start the server: clio-kit mcp-server parquet.

Check that it works

Start the server with `clio-kit mcp-server parquet`; the client's tool list should show summarize_tool, read_slice_tool, get_column_preview_tool, and aggregate_column_tool. Calling summarize_tool on any Parquet file and getting a schema and row count confirms the connection.

Troubleshooting

  1. If the server is not found, verify the server name is `parquet` (not `parquet-mcp`).
  2. For import errors, run `uv sync --all-extras --dev` in the `clio-kit-mcp-servers/parquet` directory for local development.
  3. If `clio-kit` is not found, install uv and run `uv tool install 'clio-kit==2.10.6'`, then update the shell with `uv tool update-shell`.

Things to try

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

  • Summarize this Parquet file's schema, row count, and file size.
  • Read rows 0-100 from this Parquet file, projecting only the selected columns.
  • Preview the values of the 'temperature' column with pagination.
  • Compute the min, max, and mean of this Parquet column.

Tools 4

summarize_tool read-only
Return Parquet schema, row count, and file size.
read_slice_tool read-only
Read a row slice from a Parquet file with optional column projection and filtering.
get_column_preview_tool read-only
Preview values from a specific column with pagination.
aggregate_column_tool read-only
Compute aggregate statistics (min, max, mean, etc.) on a Parquet column.

Use cases

Quickly obtain the schema, row count, and size of a Parquet file.
Read row slices from a Parquet file for inspection with optional filtering.
Preview values from a specific column to get a first look at the data.
Compute statistical summaries, like mean or max, on a Parquet column.

Supported clients

Claude Desktop
Claude Code
Cursor
VS Code

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

Overview

CLIO Parquet is an MCP server for reading and analyzing Apache Parquet files, part of the CLIO Kit, a tooling layer of the IoWarp platform. It offers tools to get file summaries, read row slices, preview column values, and compute column statistics. The server is distributed via the Python package clio-kit and communicates over stdio. It is suitable for scientific or tabular data analysis where data is stored in Parquet format, often within HPC environments.

Similar servers

Compare with other MCP servers

More servers from this repository

CLIO Arxiv MCP ServerCLIO CompressionHDF5 MCP ServerCLIO LmodCLIO Node HardwareCLIO PandasCLIO ParaViewCLIO Plot MCP ServerCLIO SeismicCLIO SlurmCLIO SpackCLIO Web MCP Server

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