Probe Helpcenter
Auto Light Dark
Auto Light Dark

Connect Claude Code to Probe

Connecting an AI assistant lets you ask questions about your boards in plain language. This guide covers Claude Code, the assistant we have tested with the evaluation server. It takes about ten minutes and is done once.

Using another assistant? See Connect another AI assistant.

Before you start

  • You have opened your invite link and can reach your welcome page. See Join your evaluation.

  • You need your own Claude account. It is not provided by E-Sharp.

Step 1: Install Claude Code

Windows (PowerShell):

PowerShell
irm https://claude.ai/install.ps1 | iex

macOS, Linux or WSL:

Bash
curl -fsSL https://claude.ai/install.sh | bash

Alternatives: winget install Anthropic.ClaudeCode on Windows, or brew install --cask claude-code on macOS.

Confirm the installation:

claude --version

Step 2: Sign in to Claude

Start Claude and follow the browser prompt to sign in:

claude

Your credentials are stored after the first login. To switch accounts later, type /login inside a running session.

Personal plans (Pro or Max): before you use real design data, check the privacy settings in your Claude account and turn off permission to use your chats for model training. Business plans (Team, Enterprise, API) exclude training by contract. See Data protection & security.

Step 3: Connect Claude to Probe

  1. On your welcome page, under Connect your AI assistant, press Create my key.

  2. The page shows two claude mcp add commands, one for probe and one for trace. Copy them right away: your key is shown only once.

  3. Paste each command into a terminal and run it.

The settings are saved and reused every time you start Claude.

  • probe gives Claude access to the board data, the review findings and the datasheets.

  • trace handles files in the background and is required for Probe to work.

Your key is personal, like a password. Each person creates their own.

Step 4: Verify

Start Claude and type:

/mcp

Both probe and trace should be listed as connected.

Step 5: Decide how Claude asks for approval

By default, Claude Code asks before every tool call, including every call to Probe. A review question can take a dozen or more calls, so you will be asked many times. How Claude asks for approval is your decision. There are three options:

Option

What happens

Our advice

1. Approve each call (default)

Claude asks before every action.

The most control, but slow for review work.

2. Allow Probe and Trace only

Calls to Probe and Trace run without asking. Claude still asks before it edits files or runs commands on your computer.

Recommended. Few interruptions, and the calls that run without asking only go to your evaluation server.

3. Skip all approvals (claude --dangerously-skip-permissions)

Claude does everything without asking, including running commands and changing files on your computer.

Anthropic recommends this only in an isolated environment, such as a container or virtual machine. You may have seen it in our demo; on your own computer, choose option 2.

To choose option 2, add this to your Claude Code settings file, ~/.claude/settings.json (on Windows %USERPROFILE%\.claude\settings.json):

JSON
{
  "permissions": {
    "allow": ["mcp__probe", "mcp__trace"]
  }
}

Or type /permissions in Claude and add mcp__probe and mcp__trace as allow rules. If your company manages Claude Code settings centrally, your IT department may decide this for you.

When your key expires

A key works for 14 days, which covers the whole evaluation. If you lose it or it expires, press Create a new key on the welcome page (the old key stops working at once). Then remove the old connections and run the new commands:

claude mcp remove probe
claude mcp remove trace

Tip: choose a strong model for review questions

Design-review questions are tool-heavy: a question like "trace this net and tell me what's connected" can take a dozen or more calls to Probe. Use the strongest model and a high effort level (/model and /effort) for review work. For quick lookups, such as a part number, a lighter setting gives the same answer with less usage. You can check your usage with /usage.