mcp MCP — Use Cases, Install & Live Demo

Why use it

Key features

Drives your existing Chrome profile — already-logged-in sessions work
Local — no external browser automation service
Uses real browser fingerprint; sidesteps most bot detection
Free (Chrome extension + local MCP)

Live Demo

What it looks like in practice

mcp.replay ▶ ready

0/0

Install

Pick your client

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

{
  "mcpServers": {
    "mcp": {
      "command": "npx",
      "args": [
        "-y",
        "mcp"
      ],
      "_inferred": true
    }
  }
}

Open Claude Desktop → Settings → Developer → Edit Config. Restart after saving.

~/.cursor/mcp.json · .cursor/mcp.json

{
  "mcpServers": {
    "mcp": {
      "command": "npx",
      "args": [
        "-y",
        "mcp"
      ],
      "_inferred": true
    }
  }
}

Cursor uses the same mcpServers schema as Claude Desktop. Project config wins over global.

VS Code → Cline → MCP Servers → Edit

{
  "mcpServers": {
    "mcp": {
      "command": "npx",
      "args": [
        "-y",
        "mcp"
      ],
      "_inferred": true
    }
  }
}

Click the MCP Servers icon in the Cline sidebar, then "Edit Configuration".

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "mcp": {
      "command": "npx",
      "args": [
        "-y",
        "mcp"
      ],
      "_inferred": true
    }
  }
}

Same shape as Claude Desktop. Restart Windsurf to pick up changes.

~/.continue/config.json

{
  "mcpServers": [
    {
      "name": "mcp",
      "command": "npx",
      "args": [
        "-y",
        "mcp"
      ]
    }
  ]
}

Continue uses an array of server objects rather than a map.

~/.config/zed/settings.json

{
  "context_servers": {
    "mcp": {
      "command": {
        "path": "npx",
        "args": [
          "-y",
          "mcp"
        ]
      }
    }
  }
}

Add to context_servers. Zed hot-reloads on save.

claude mcp add mcp -- npx -y mcp

One-liner. Verify with claude mcp list. Remove with claude mcp remove.

Use Cases

Real-world ways to use mcp

Automate tasks behind your real web logins (Gmail, admin panels)

👤 Productivity power users, ops ⏱ ~15 min intermediate

When to use: A service has no API and you're logged in via SSO — you want Claude to do repetitive work in that tab.

Prerequisites

Chrome + Browser MCP extension installed — docs.browsermcp.io → install extension

Flow

Grant the extension access to the target tab

Open the admin panel tab, click the Browser MCP extension icon, allow this tab.✓ Copied

→ Extension shows 'connected'
Delegate a task

In the admin panel: find the user with email X, change their plan to 'Pro', take a screenshot of the confirmation.✓ Copied

→ Task completes with visual proof

Outcome: Repetitive admin actions done via chat, with zero API integration work.

Pitfalls

Running actions you didn't mean — this is YOUR real account — Be precise in prompts; the MCP doesn't undo. Test on staging accounts first

Extract data from sites that block headless scrapers

👤 Researchers, analysts ⏱ ~15 min intermediate

When to use: A site you're allowed to read programmatically blocks Playwright/Selenium fingerprints.

Flow

Navigate manually first if needed

Browser MCP uses your open tabs; navigate to the target, then ask me to read it✓ Copied

→ Correct tab is active
Extract structured data

Read the results table on this page and return as JSON with columns [name, date, value]✓ Copied

→ Structured data

Outcome: Data you couldn't get with headless tools, obtained via your normal browser session.

Pitfalls

Using this to scrape at scale against a site's ToS — Stay within rate limits a human would generate; respect robots.txt

Combine with: filesystem

Reproduce a user bug in your real browser

👤 Support engineers ⏱ ~10 min beginner

When to use: A customer reports a bug; you want to replicate their exact click path in your own logged-in session to see.

Flow

Replay the reported steps

Reproduce this user's flow in the app tab: login → Settings → toggle X → save. Report what happens and any console errors.✓ Copied

→ Reproduction result with errors

Outcome: Faster support triage than video back-and-forth.

Combine with: sentry

Combinations

Pair with other MCPs for X10 leverage

mcp + filesystem

Save scraped data to disk

Extract the table on this page and save as ./exports/{hostname}-{date}.csv.✓ Copied

mcp + sentry

Reproduce a user bug tied to a Sentry event

Sentry event XYZ shows breadcrumbs [open, click, save]. Replay them in my current tab and compare the result.✓ Copied

Tools

What this MCP exposes

Tool	Inputs	When to call
browser_navigate	url: str	Move the current tab
browser_snapshot		See structured page content
browser_click	ref or selector	Click elements
browser_type	ref, text	Fill inputs
browser_screenshot	fullPage?	Capture state

Cost & Limits

What this costs to run

API quota: None — local
Tokens per call: Snapshots 500-3000
Monetary: Free
Tip: Prefer snapshot-based refs over screenshots for action — cheaper and more reliable

Security

Permissions, secrets, blast radius

Credential storage: Uses your real browser's cookies — these are as sensitive as your sessions

Data egress: Tabs connect to whatever you navigate to; no telemetry

Troubleshooting

Common errors and fixes

Extension shows 'not connected'

Click the extension icon in the target tab and allow; the connection is per-tab

Verify: Badge shows green

Actions do nothing on an SPA

Take a snapshot first — SPAs often need a wait before elements are interactive

Site detected automation anyway

Rare — disable other extensions that modify navigator properties; restart browser

Alternatives

mcp vs others

Alternative	When to use it instead	Tradeoff
Playwright MCP	You want hermetic, scriptable, CI-friendly browser automation	No real-session access; more bot-detectable
Puppeteer MCP	Chromium-only headless control	Same tradeoffs as Playwright, smaller surface

More

Resources

📖 Read the official README on GitHub

🐙 Browse open issues

🔍 Browse all 400+ MCP servers and Skills