Experiment tracking for Claude Code, Codex and Cursor
Probe is an experiment tracker that coding agents use to log, share, and access runs, metrics, and decisions in a single record for teams and agents.
useprobe.aiCrawled Nov 15, 2026
Probe is an experiment tracker that coding agents use to log, share, and access runs, metrics, and decisions in a single record for teams and agents.
Probe’s hosted MCP server allows Claude Code, Codex, Cursor, and other MCP clients to access team projects, experiments, runs, and metrics with OAuth sign-in and read-only tools.
Track reinforcement learning runs with Probe, including rewards per step, rollouts, turns and tool calls as spans, and Harbor trials, with support for Miles, TRL, verl, SkyRL and ART.
When many coding agents run ML experiments in parallel, Probe shows which branch is winning, what each branch cost in agent-hours, and lets every agent start from what the others already tried.
The guide explains how to use Claude Code, Codex, or Cursor for training runs, evaluations, and sweeps, with each run, question, and result recorded for future reference.
Probe tracks runs and metrics logged to Weights & Biases (W&B), imports W&B history, and supports coding agents in setting up, writing to, and reading from the system.
MLflow is an open-source platform for the machine learning lifecycle. Probe is a hosted experiment tracker built for coding agents. How they differ and when to pick each.
Trackio is a free, lightweight, local-first experiment tracker from Hugging Face with a wandb-compatible API. Probe is a hosted team record built for coding agents. How they differ.
Experiment tracking for coding agents. Claude Code, Codex and Cursor log your runs, metrics and decisions, so agents build on what your team already tried.
How Probe, Corp. collects, uses, shares, retains, and protects personal data and Customer Data in the Probe platform.