# Oxynet

> A computational layer for cardiopulmonary exercise testing (CPET). Oxynet reads the
> raw export of a metabolic cart (23 file formats, detected automatically) and
> returns ventilatory thresholds (VT1 and VT2), substrate use (fat and carbohydrate
> oxidation, FATMAX), exercise oscillatory ventilation and derived CPET quantities as
> structured, versioned JSON. Research software, not a diagnostic device.

Oxynet is reached three ways: a REST API, a remote MCP server for AI assistants, and
the web app at https://app.oxynet.net. All three use the same file parsers.

Results carry no confidence score, because nothing is calibrated against clinical
outcomes. Oscillatory ventilation analysis is beta: it was developed on one heart
failure cohort and its transportability to other populations is untested.

## If you are an AI agent

- [Agent guide](https://app.oxynet.net/llms.txt): connect, upload a file, analyse it. Start here.
- [Full agent guide](https://app.oxynet.net/llms-full.txt): the whole API in one text file.
- [OpenAPI 3.0 schema](https://app.oxynet.net/v1/openapi.json): importable as tools or actions.
- [MCP server](https://app.oxynet.net/oxynet-mcp): remote MCP over HTTP, with an API key or OAuth.
- [Oxynet in Claude's connector directory](https://claude.ai/directory/oxynet): for people using Claude on the web, Desktop or mobile. Connect there and sign in once with the API key.
- [Official MCP Registry entry](https://registry.modelcontextprotocol.io/v0.1/servers?search=net.oxynet/cpet): `net.oxynet/cpet`, for clients that install servers from the registry.
- [Synthetic CPET](https://app.oxynet.net/v1/sample): demonstrate the API without a patient file.
- [Free API key](https://app.oxynet.net/signup): if the user has no key, send them here. A person signs in with Google or GitHub; an agent cannot get a key itself.
- [Agent skill](https://github.com/andreazignoli/oxynet-skills): the same instructions packaged for Claude Code, Gemini CLI, Codex and other coding agents.

Upload the vendor file exactly as exported. Do not parse it or convert its units
yourself: Oxynet knows each format's units and clock conventions. Do not estimate a
threshold from the trace either. The assistant routes the question and explains the
answer; Oxynet's validated models decide the physiology.

## If a person asks how to find VT1 and VT2 in their file

- [VT1 and VT2 from a CPET file](https://www.oxynet.net/ventilatory-thresholds): the three ways to run it, what comes back, who decides the threshold, common questions.
- [VT1 and VT2 from a COSMED Omnia export](https://www.oxynet.net/ventilatory-thresholds/cosmed): which COSMED exports are read and what the parser handles in them.
- [VT1 and VT2 from a CORTEX export](https://www.oxynet.net/ventilatory-thresholds/cortex): which CORTEX exports are read and what the parser handles in them.

## Start

- [Overview](https://www.oxynet.net/developers)
- [Quick start](https://www.oxynet.net/developers/quickstart)
- [Downloads](https://www.oxynet.net/developers/downloads)

## Interfaces

- [REST API (v1)](https://www.oxynet.net/developers/api/v1)
- [Python](https://www.oxynet.net/developers/python)
- [MCP and AI agents](https://www.oxynet.net/developers/mcp)

## Reference

- [Supported formats](https://www.oxynet.net/developers/formats)
- [Analysis catalogue](https://www.oxynet.net/developers/analyses)
- [Outputs and provenance](https://www.oxynet.net/developers/outputs)
- [Data handling](https://www.oxynet.net/developers/data-handling)

## Partners

- [Integration patterns](https://www.oxynet.net/developers/patterns)
- [Partner onboarding](https://www.oxynet.net/developers/partners)

## Peer-reviewed publications

- [AI-Driven Analysis of CPET to Identify Gas Exchange and Ventilatory Thresholds](https://pubmed.ncbi.nlm.nih.gov/41784915/): Sports Medicine, 2026
- [AI for CPET Interpretation](https://www.sciencedirect.com/science/article/abs/pii/S1746809423002690): Biomedical Signal Processing and Control, 2023
- [AI Technologies in Exercise Data Processing](https://link.springer.com/article/10.1007%2Fs11332-019-00557-x): Sport Sciences for Health, 2019
- [LSTM Networks for VO₂ Estimation](https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0229466): PLOS ONE, 2020
- [LSTM for Intensity Domain Estimation](https://doi.org/10.1080/17461391.2019.1587523): European Journal of Sport Science, 2019
- [Crowdsourcing and CNN for Intensity Domain Determination](https://doi.org/10.1080/17461391.2020.1866081): European Journal of Sport Science, 2022
- [Regression, Generation, and Explanation](https://doi.org/10.3390/s23020826): Sensors (MDPI), 2023

## Optional

- [Python package (pyoxynet)](https://pypi.org/project/pyoxynet/): the open-source research package. It is separate from the hosted API.
- [User manual (PDF)](https://www.oxynet.net/manual)
- [Integration](https://www.oxynet.net/integration): Oxynet inside a data pipeline or a product.
- [Demo](https://www.oxynet.net/demo)