MCP package / Quick start

Get Serge MCP running
in about 10 minutes

This guide gets you from zero to your first local report: install the package in Claude Desktop, run a benchmark prompt, inspect the output, and re-run after fixes.

Need

Node.js 18+ and Claude Desktop with an active Claude subscription.

Stores

Session data and HTML reports under ~/.serge/.

Best for

Offline testing, agent journey debugging, and showing the concept quickly.

1

Add Serge MCP to Claude Desktop

Open your Claude Desktop config file:

macOS ~/Library/Application Support/Claude/claude_desktop_config.json
Windows %APPDATA%\Claude\claude_desktop_config.json

Add Serge to the mcpServers object:

2

Restart Claude Desktop fully

Quit the app completely, not just the window. Then reopen it. On the first run, Serge MCP installs the browser dependency it needs. That can take up to a minute.

If the tools loaded correctly, you should see Serge tools in the Claude Desktop tool picker.

3

Run your first benchmark

Open a fresh Claude conversation and use a concrete prompt:

Keep prompts task-shaped. Product search, add-to-cart, navigation, support, and form flows all work well.

Use Serge to search for running shoes on yourstore.ch and add the first result to the cart.

Use Serge to find the contact form on yourstore.ch and fill it with test data.

Use Serge to navigate from the homepage to the return policy on yourstore.ch.

4

Read the local report

When the run ends, Serge opens a local HTML report in your browser. It includes the full action timeline, screenshots, findings, and outcome.

You can reopen the latest report from the terminal any time:

What to look for

SummaryTask, domain, result, total time, and headline stats.
TimelineEvery step Claude took, plus screenshots and failures.
FindingsMissing labels, duplicate names, slow pages, blocked flows, inaccessible controls.
NetworkRequests, heavy assets, and the slowest resources from the run.
5

Fix, re-run, and compare

After you apply fixes, run the same prompt again and compare the outcome. To browse old runs:

Reports live under ~/.serge/reports/ and raw session data under ~/.serge/sessions/.

Fix findings with AI

Paste the findings from your report into Claude, Cursor, or your coding agent with a remediation prompt like this:

Next steps

→ Back to MCP overview→ Open the GitHub repository→ Run the hosted scanner