PyOD
Open-source Python anomaly detection library with a shared detector API, benchmark-backed orchestration, and a BSD 2-Clause license.
PyOD is a Python library for detecting unusual records and behavior in data on your own machine or server. A shared fit/predict interface lets developers and data scientists compare classical and deep learning anomaly detectors without changing the surrounding application for each algorithm. Its code uses the BSD 2-Clause license.
The detector inventory covers tabular data, time series and graphs, with embedding-based methods for text, images and audio. Methods include kNN, LOF, HBOS and COPOD alongside neural approaches. Each detector has its own assumptions and dependencies; a high anomaly score still needs interpretation in the context of the dataset.
PyOD 3 adds ADEngine for lifecycle orchestration and benchmark-informed detector selection. The od-expert skill supplies a multi-turn investigation workflow through a compatible coding agent. An optional MCP server exposes stateless tools for detector information, planning, running detection and analyzing results. Stateful investigate/iterate MCP tools remain deferred. Agent-assisted workflows depend on the chosen agent and model provider; the standard detector API runs independently.
Install the core package with pip install pyod. Use the mcp extra for the MCP server. Saved models can be reloaded for later inference, and the classic detector interface remains backward compatible.
