接入你的智能体
從發布者安裝。不是 PlugMCP 端點。
遠端 URL 還是本機指令? 兩種安裝方式的差異
A native desktop application for viewing, editing, and querying data files. Octa opens a file in a spreadsheet-like table with sorting, filtering, and search, and runs as a single binary on Linux, macOS, and Windows.
Files are not read-only: edit cells, rows and columns with undo, then save in the original format or convert to another one with Save As. Any open table, a plain CSV included, can also be queried with SQL through DuckDB.
📚 Documentation: https://thorstenfoltz.github.io/octa/

The user guide is sorted by task, and each topic lists where a job lives: in the app, on the command line, and as an MCP tool for AI agents.
Parquet, CSV/TSV, JSON and JSON Lines, Excel, ODS, Arrow, Avro, ORC, SQLite, DuckDB, GeoPackage, SAS, SPSS, Stata, R, HDF5, NetCDF, NumPy, MessagePack, BSON, DBF, XML, TOML, YAML, Jupyter notebooks, Markdown, HTML, EPUB, PDF tables, log files, GeoJSON, Shapefile, Delta Lake, Apache Iceberg, fixed-width text, zip/tar archives, SQL dumps, and source code. Most of those can be written back, and Save As converts between them.
See supported formats
for the full list and what each one can do, or open samples/ for
one small example of every one of them.
Building from source works on all three platforms: install
Rust, clone this repository, and run
cargo build --release. Linux and macOS need a few native libraries first,
Windows needs a C++ toolchain; the
installation guide
lists them per platform, along with uninstall steps and the first-launch
security prompts on Windows and macOS.
On Arch and Arch-based distributions (Manjaro, EndeavourOS, Garuda, ...) Octa is
on the AUR, either built from source or as the prebuilt binary. Both install the
man page, so man octa works:
Build from source:
paru -S octa
Or take the prebuilt binary:
paru -S octa-bin
Swap paru for yay or any other AUR helper you use.
Install the latest release with one command, user-local (no sudo):
curl -fsSL https://raw.githubusercontent.com/thorstenfoltz/octa/master/get-octa.sh | bash -s -- ~/.local
Or system-wide into /usr/local:
curl -fsSL https://raw.githubusercontent.com/thorstenfoltz/octa/master/get-octa.sh | sudo bash
On WSL, also install a file dialog backend, or Open and Save as have nothing to show and appear to do nothing (Octa says so in the status bar):
sudo apt install xdg-desktop-portal-gtk # or: sudo apt install zenity
An AppImage and a plain tarball are on the releases page. A downloaded AppImage still needs its execute bit, and a file manager will silently ignore a double-click until it has one:
chmod 750 Octa-*-x86_64.AppImage
./Octa-*-x86_64.AppImage
A headless image ships the CLI and the MCP server:
docker pull ghcr.io/thorstenfoltz/octa:latest
Mount a directory and pass any CLI flag:
docker run --rm -v "$PWD:/data" ghcr.io/thorstenfoltz/octa --schema /data/file.parquet
See the container docs for the MCP invocation and Podman.
Install from the Microsoft Store: one click, no SmartScreen prompt, and the Store keeps it updated.
Or download octa.exe from the
releases page and run it, no
installation needed. install.ps1 (per-user, no admin) or install.bat
(system-wide) set it up properly with a Start Menu entry.
Those downloads are not code-signed, so Windows shows "Windows protected your PC" on first launch. It is safe to continue: how to get past the SmartScreen prompt.
One line, no password, Apple Silicon:
curl -fsSL https://raw.githubusercontent.com/thorstenfoltz/octa/master/get-octa.sh | bash
It puts Octa.app into ~/Applications and octa into ~/.local/bin, and
because curl marks nothing as quarantined, macOS opens it without a
Gatekeeper warning. Intel Macs build from source.
The Octa.app bundle is also on the
releases page. It is not
notarised, so a copy downloaded with a browser makes macOS claim "the developer
cannot be verified" (and occasionally that the app is damaged). Neither means
anything is wrong with the download:
how to open it anyway.
MIT
Fast native desktop app for exploring 20+ data formats like Parquet, DuckDB, CSV, Excel, Arrow, GeoJSON, Jupyter and more, plus connections to Postgres, Snowflake, BigQuery, GCS, S3 and more. DuckDB S
總覽
Fast native desktop app for exploring 20+ data formats like Parquet, DuckDB, CSV, Excel, Arrow, GeoJSON, Jupyter and more, plus connections to Postgres, Snowflake, BigQuery, GCS, S3 and more. DuckDB SQL, charts, schema diffs, and an MCP server so AI agents query the same data.
工具
尚無工具清單。遠端伺服器會在首次健康檢查後取得工具清單。
常見問題
在本頁複製適用於 Cursor、Claude Desktop、Claude Code 或 VS Code 的 stdio 設定。它使用發布者的安裝方式,而不是 PlugMCP 代理伺服器。
這筆記錄未列出必要的環境變數。
Octa 透過 stdio 在本機執行。
上次檢查時間:2026-11-15。 Fast native desktop app for exploring 20+ data formats like Parquet, DuckDB, CSV, Excel, Arrow, GeoJSON, Jupyter and more, plus connections to Postgres, Snowflake, BigQuery, GCS, S3 and more. DuckDB SQL, charts, schema diffs, and an MCP server so AI agents query the same data.