MCP package · local-first

Serge MCP for
Claude Desktop

Serge MCP is the local package teams can use to test flows offline before wiring the full Serge product. Claude drives the browser, Serge captures the steps, and the report stays on your machine as a self-contained HTML artifact.

Quick start guideGitHubScanner API docs
Jump
Test it in minutes

Add one Claude Desktop config block, restart the app, and run your first benchmark prompt.

Jump
Keep everything local

Reports, screenshots, and session history stay under ~/.serge on the test machine.

Jump
Understand the toolset

Serge MCP gives Claude explicit browse, read, click, type, scroll, screenshot, and report tools.

Quick install

Add Serge MCP to Claude Desktop and run it locally with no Serge account, no API key, and no extra billing. It rides on your existing Claude Desktop subscription.

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

Restart Claude Desktop, then start with a prompt like:

The first run installs the browser dependency automatically. When the task finishes, Serge opens a local HTML report in your browser.

What you get

Step timelineEvery navigation, read, click, type, and retry is preserved so you can see where the agent got stuck.
ScreenshotsEach step includes visual proof, which makes debugging much faster than reading a raw trace.
FindingsSerge flags missing labels, bot blocks, slow loads, inaccessible controls, duplicate names, and other issues that break agent journeys.
Offline artifactThe report is a local HTML file you can open, archive, or send to a teammate without a special viewer.

Where files live

Serge MCP stores session artifacts locally:

That makes it useful for internal testing, design reviews, and quick agent-flow debugging before you commit to the hosted product.

Example prompts

Copy these directly into Claude Desktop. In practice, telling Claude to “Use Serge” keeps it on the MCP tools instead of drifting to the built-in browser.

Use Serge to find the cheapest laptop under 1000 CHF on yourstore.ch and add it to the cart.

Use Serge to search for running shoes on yourstore.ch and add the first in-stock 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.

Tools Claude receives

Serge MCP exposes eight tools to Claude Desktop so the session is structured and measurable instead of opaque.

serge_start_session

Start a benchmark session with a target domain and a task description.

serge_navigate

Open a URL and return the page title plus accessibility tree context.

serge_read_page

Re-read the current page after scrolling, waiting, or dynamic changes.

serge_click

Click an element by role and accessible name so the action remains machine-readable.

serge_type

Type into an input or search field, optionally submitting with Enter.

serge_scroll

Move through the page to reveal content below the fold or trigger lazy-loaded sections.

serge_screenshot

Capture a screenshot for visual debugging when the accessibility tree is not enough.

serge_end_session

End the run, generate the report, and record the outcome.

Useful commands

Once the package is installed through Claude Desktop, you can open reports and review history from the terminal:

How it fits with Serge today

Serge MCP is the fastest local way to trial the core idea: can an AI agent complete a real task on this site?

The hosted Serge product goes further with passive agent traffic detection, session visibility, replay orchestration, and team-shared reporting. The MCP package is still useful as the offline, zero-account quick start.

Next

→ Open the step-by-step quick start→ Open the Scanner API docs→ Run the public scanner