For imaging scientists and research software teams

Turn high-content microscopy images into reproducible measurements.

OpenHCS is designed for high-content studies spanning many wells, sites, channels, Z planes, or time points. It keeps source selection, processing steps, and result definitions in one validated pipeline across the desktop GUI, Python, supported CellProfiler .cppipe imports, and local agents.

Install OpenHCS 0.8.1 Is OpenHCS for my data?

Application gallery

OpenHCS in use

Real OpenHCS sessions showing pipeline editing, execution, images, and results. Pointer-free clips are driven through the local MCP; open any still or poster at full resolution.

Review the uncut agent evidence

OpenHCS Plate Manager with seven assay plates loaded, including CellProfiler Comet Assay and Wound Healing examples — Workspace
Seven assay plates, one workspace

Five deterministic examples and two imported CellProfiler workflows retain independent state in one session.

OpenHCS Pipeline Editor showing the 12 imported steps of the CellProfiler Comet Assay — Authoring
CellProfiler steps in the same pipeline model

The 12-step Comet Assay uses OpenHCS compilation, configuration, viewers, and results.

OpenHCS workspace beside pipeline and step settings showing Well Filter Image30 at pipeline scope and inherited by the step — Configuration
See configuration provenance in both directions

Well Filter changes to Image30 in PipelineConfig, then the inherited step field flashes as the update arrives. Clicking that step label flashes the owning PipelineConfig field, making the provenance link explicit in both directions.

Fiji ImageJ showing one NeuronCyto II Field 1 nuclear plane with its nine native segmentation ROI entries — Fiji
Review matching ROIs in Fiji

OpenHCS streams one bound Field 1 nuclear plane and its nine segmentation ROIs together. The native outlines follow the displayed nuclei.

OpenHCS showing an execution endpoint preparing its function catalog while a pipeline compiles — Local MCP
See endpoint startup while the request is running

A pointer-free MCP compile request immediately adds the execution endpoint, streams catalog-preparation phases into the browser and status bar, and settles connected.

Napari showing segmented nuclei, a selected native ROI, its measurement row, and the current Z plane — Napari
ROI selection navigates the Z stack

A pointer-free MCP sequence selects measurement rows across a three-plane result. Napari scrolls the table, highlights the native ROI, and moves from Z 3/3 to Z 1/3 and back.

Pipeline representation

The same pipeline in the UI, Python, headless runs, and MCP

UI edits and generated Python use the same typed pipeline, function, and artifact model.

UI and Python round trip

Build visually, edit generated Python, and inspect typed artifacts without translating between pipeline formats.

CellProfiler `.cppipe` import

Import supported CellProfiler modules as regular OpenHCS steps; unsupported modules fail clearly.

CellProfiler API

Plane-local and volumetric analysis

Dimensionality is function-defined, not a global 2D/3D switch. Current volumetric routes include Watershed segmentation, 3D intensity, size and shape, and occupied-volume measurements; plane-local labels are not silently stitched across Z.

Capability reference

View images, ROIs, and measurements

Send compiled pipeline results to Napari or Fiji.

napari

Fiji openhcs[gui,viz]

GPU libraries + custom functions

Decorate ordinary Python functions for CuPy, PyTorch, JAX, TensorFlow, or pyclesperanto. They gain contract validation, UI integration, and automatic memory conversion.

openhcs[gpu]

Compute with

CuPy

PyTorch

JAX

TensorFlow

pyclesperanto Your functions

Bio-Formats image I/O

Read microscopy images and dimensional metadata through the Bio-Formats backend.

Bio-Formats openhcs[bioformats]

Agent workflow

Use OpenHCS through a local agent

Give the agent a microscopy folder or plate and the analysis you want. Through MCP it can construct and validate the same typed pipeline used by the desktop application, run it locally, and inspect the results without manually constructing the pipeline. Every pipeline remains editable in the GUI and as generated Python.

Install local MCP View current releases

Archived model run

One prompt, two steps, complete result

Starting from one prompt and a public NeuronCyto II crossover field, Codex projected two loose TIFF planes into biological channels, built and compiled a pipeline in the live desktop, and ran a per-neuron morphology analysis. The unattended run produced persistent images, ROIs, SWC morphology, measurements, and a settled Napari result table. Later human visual QA exposed a clustered crossover-tracing defect; after the generic tracing fix, a separate recapture showed 9 neurons and 24 spatial-graph paths. The generated Python remained editable in OpenHCS.

Client: Codex 0.146.0
Model: gpt-5.6-sol
OpenHCS: 0.7.13 at f1c1d9b67
Intervention: One prompt; no later human steering

From source binding to per-neuron morphology A 2:38 showcase: the first 2:20 is an edit of the original unattended 10:47 run, from function discovery through the editable two-step pipeline, compilation, neurite graph, ROI manager, and measurements. A clearly labelled post-run replay then cycles five neuron selections while the linked neurite overlay and ROI rows change together. The original run produced 9 neurons and 25 spatial-graph paths.

Uncut run and machine-verifiable records

The complete 10:47 recording remains the visual authority for the unattended run. The first 2:20 above contains only excerpts from that recording; the clearly labelled final segment is a separate post-run MCP interaction replay, not evidence of actions taken by the unattended agent.

The client had no shell or repository access. Mutating operations were pre-authorized inside the isolated recording session; this is not a claim about every registered client or model. The run exposed an SWC inventory omission, corrected at the generic format owner in 973c51fd0. Subsequent visual QA exposed the crossover defect, corrected at the registered tracing strategy in 0eb5f77c0.

Installation

Desktop installers include the GUI and local MCP setup. Use PyPI for headless systems, GPU libraries, or an existing Python environment.

Desktop installers No Python setup required

GitHub release

Download for WindowsDownload and run — no ZIP to extract

Download for macOSOpen the DMG, then open OpenHCS Installer

User-scoped, CPU-only installers include CellProfiler compatibility, local MCP, Napari, Fiji, and Bio-Formats. GPU libraries are optional and not included; Fiji downloads Java on first use. The latest complete GitHub release may trail PyPI. The Windows installer is unsigned, and the macOS installer is not notarized, so the OS may ask you to confirm trust.

Python and headless installs For existing Python environments

PyPI

$ python -m pip install "openhcs[gui,viz,bioformats,mcp,cellprofiler-compat]"
$ openhcs

Desktop onlyopenhcs[gui]
Headlessopenhcs
CUDA opt-inopenhcs[gui,gpu]

View OpenHCS on PyPI

PyPI 0.8.1Python 3.11+Windows · macOS · LinuxMIT licensed

Turn high-content microscopy images into reproducible measurements.

OpenHCS in use

Seven assay plates, one workspace

CellProfiler steps in the same pipeline model

See configuration provenance in both directions

Review matching ROIs in Fiji

See endpoint startup while the request is running

ROI selection navigates the Z stack