An Agent-Computer Interface for Machine Learning

Data Science doesn't need a better model.
It needs a better Harness.

An MCP server that turns Claude into a disciplined data scientist. Structured tools, enforced methodology, real-time observability.

See it in action

Harness Studio is your window into your agent's mind

A companion dashboard that runs alongside your agent, giving you full observability into every decision it makes — live, in real time.

Dashboard

Everything at a glance

Project vitals, experiment verdict breakdown, primary metric trend with error bars, live MCP activity feed, and a mini pipeline DAG — all updating as your agent works.

Dashboard

Activity Feed

Watch it think

Every MCP tool call streamed live with full parameters and results. You see exactly what your agent is doing and why — no black box.

Activity Feed

Pipeline DAG

See the full picture

Interactive pipeline topology. Click any node for full config details. Models added by experiments show with dashed borders and EXP badges.

Pipeline DAG

Experiments

Watch it learn

Every experiment with its hypothesis, verdict, and metric deltas. The trend chart tracks your primary metric across iterations — you can literally watch your agent get smarter.

Experiments

Diagnostics

Go deep when you need to

Per-run deep dive: headline metrics, meta-learner coefficients, model correlation heatmap, calibration curves, and per-fold breakdown.

Diagnostics

Data Sources

Know what went in

Overview of ingested data, registered features, and view definitions. Full transparency into the data pipeline.

Data Sources

Train your first model in 60 seconds

Just tell Claude what you want to predict.

# In Claude Code
/plugin marketplace add msilverblatt/harness-ml
/plugin install harnessml@msilverblatt-harness-ml
View on GitHub