# Run your first review

> The review runs in your AI tool over MCP; Probe presents the cited findings in the browser.

_Source: E-Sharp (esharp.se) · Probe guide_


The review runs with AI in your own tool. Probe itself holds no generative model — it presents the results and the evidence in the browser.

## Connect your AI tool

1. Connect your assistant (Claude or Copilot) to Probe over **MCP** (Model Context Protocol) with your personal key.
2. The assistant now asks the actual board through tool calls instead of answering from memory.

## Run the review

1. Point the assistant at the design and ask it to review it.
2. Probe reads the layout, schematic and each component's datasheet together, and checks what was built against what each part requires.
3. The findings are written back to Probe, each tied to net, layer, component and datasheet page.

## Three modes per check

Probe distinguishes **checked and clean**, **couldn't be checked** (missing data, named) and **not yet reviewed**. A rule without enough data abstains rather than passes — so a clean report means reviewed, not guessed.

Open the findings list in the browser to [read and export the findings](/en/probe/support/read-and-export-findings). Just want to try without setting anything up? [Book a session](/en/book-demo) on an evaluation board.
