Blueprint MCP for Edge

Give Claude Code (and other AI assistants) direct control of your real Edge browser

Stop wrestling with headless browsers, bot detection, and authentication flows. Blueprint MCP for Edge bridges the Model Context Protocol (MCP) with your actual Edge browser - keeping you logged into all your sites, preserving your extensions, and using your real browser fingerprint.

THE CRITICAL ADVANTAGE: Efficient browser automation that does not consume your AI's context. Instead of sending massive page snapshots with every action (which quickly exhausts context limits and often exceeds MCP response size limits), Blueprint MCP lets your AI interact directly - click, type, navigate - and only fetch page content when specifically needed. This enables complex, multi-step workflows that would be impossible with snapshot-based approaches.

Perfect for developers using Claude Code, Claude Desktop, Cursor, or any MCP-compatible AI client who need to automate web tasks without the headaches of Playwright/Puppeteer/Selenium.

WHY BLUEPRINT MCP + CLAUDE CODE CHANGES EVERYTHING

The problem: Traditional browser automation requires sending full page snapshots after every action, consuming thousands of tokens per interaction and hitting response size limits. You also lose logged-in sessions, face bot detection, and spend time configuring browser profiles.

The solution: Blueprint MCP connects your AI assistant to your real Edge browser through direct commands. Your AI controls the browser - click, type, navigate - without needing page snapshots unless explicitly requested. This preserves context for actual decision-making instead of wasting it on redundant page content.

What this enables:
• AI-powered web scraping that bypasses bot detection
• Automated form filling with your saved passwords
• Testing workflows using your logged-in accounts
• Content extraction from authenticated sites
• Browser automation that works with 2FA and CAPTCHAs

KEY FEATURES

Real Browser Automation:
• Uses your actual Edge profile - Stays logged into Gmail, GitHub, AWS, etc.
• Stealth mode - No bot detection. Uses your real browser fingerprint
• Extension support - Works alongside your existing Edge extensions
• Session persistence - No re-authentication needed

Native MCP Integration:
• 20+ browser tools exposed via Model Context Protocol
• Tab management - Create, switch, close tabs programmatically
• DOM inspection - Get accessible page content
• Network monitoring - Capture requests, replay with authentication
• JavaScript execution - Run code in page context

Developer Experience:
• Zero setup - Install extension, works instantly
• Auto-reconnection - Survives browser restarts
• Debug mode - Verbose logging for troubleshooting
• Status visibility - Real-time connection status

Two Connection Modes:
• Free (Local) - WebSocket on localhost:5555, no cloud dependencies
• PRO (Cloud Relay) - Control browser from anywhere, multiple AI clients

Comprehensive Automation:
• Screenshots (full page or elements)
• Content extraction as Markdown
• Form automation (fill, dropdowns, file uploads)
• Mouse/keyboard control (click, type, hover, drag-drop)
• Performance metrics (Web Vitals)

Advanced Capabilities:
• Tech stack detection (40+ frameworks)
• Network request replay
• PDF export
• Dialog handling (alert/confirm/prompt)
• Extension management

HOW IT WORKS

Architecture:
AI Assistant (Claude Code) -> MCP Protocol -> MCP Server -> JSON-RPC over WebSocket -> Edge Extension (this) -> Chrome APIs -> Your Edge Browser

Setup (3 steps, 2 minutes):
1. Install MCP server: npm install -g @railsblueprint/blueprint-mcp
2. Install this Edge extension
3. Configure AI client: claude mcp add browser npx @railsblueprint/blueprint-mcp@latest

The extension auto-connects when your AI client starts the MCP server.

USE CASES

1. Authenticated Web Scraping
Before: Manually log in, navigate, copy-paste data. Bot detection, session timeouts.
With Blueprint MCP: "Go to dashboard.company.com and extract the Q4 metrics table" - Uses your logged-in session automatically.

2. Multi-Step Form Automation
Before: 2 hours/week manually filling vendor forms across 5 portals.
With Blueprint MCP: "Fill out the vendor forms with my company info" - 2 hours -> 5 minutes.

3. Testing Logged-In Workflows
Before: Create test accounts, configure webhooks, manual testing.
With Blueprint MCP: "Test the upgrade flow from free to pro" - Tests with real authentication.

4. Documentation Screenshots
Before: 3+ hours manually capturing and editing screenshots.
With Blueprint MCP: "Generate screenshots for all onboarding steps" - Consistent screenshots in minutes.

5. API Testing via Browser
Before: Copy cookies manually, troubleshoot auth, replay requests.
With Blueprint MCP: "Capture API calls when I click Submit and replay them" - Debug with actual browser authentication.

TECHNICAL DETAILS

Available MCP Tools (20+):
Connection: enable, disable, status, auth, browser_connect
Tabs: browser_tabs (list/new/attach/close)
Navigation: browser_navigate (url/back/forward/reload)
Content: browser_snapshot, browser_take_screenshot, browser_extract_content, browser_console_messages, browser_network_requests
Interaction: browser_interact (click/type/hover/scroll), browser_fill_form, browser_lookup, browser_drag
Advanced: browser_evaluate, browser_handle_dialog, browser_window, browser_pdf_save, browser_performance_metrics, browser_verify_text_visible, browser_verify_element_visible
Extensions: browser_list_extensions, browser_reload_extensions

API Capabilities:
• Accessible DOM snapshots without screenshots
• CSS selectors + :has-text() pseudo-selector
• Network monitoring with filters (URL, method, status, type)
• JSONPath filtering for large JSON responses
• JavaScript execution with async/await support
• Chrome DevTools Protocol (CDP) access

Limitations:
• Cannot automate edge:// or chrome:// pages (Browser restriction)
• Screenshot size limit: 10MB
• Network history limit: 1000 requests

Requirements:
• Edge 90+ (Chromium-based)
• Node.js 18+ (for MCP server)
• MCP-compatible AI client
• Port 5555 available (Free tier)

SECURITY & PRIVACY

Free Tier (Local Mode):
• All communication stays on your machine (localhost WebSocket)
• Zero data sent to cloud
• No tracking, telemetry, or analytics
• Works completely offline
• No account required

PRO Tier (Cloud Relay):
• Commands and responses pass through relay (including page content/screenshots when requested)
• OAuth2 tokens stored locally (encrypted by Edge)
• No logging - data passes through relay but is not stored or logged anywhere
• Your browsing activity is not tracked or monitored

Permissions Explained:
debugger - Chrome DevTools Protocol for screenshots, DOM snapshots, network monitoring
tabs - Tab management (list, create, switch, close)
activeTab - Current tab access for content reading
storage - Local preferences and OAuth tokens (PRO)
alarms - Auto-reconnection and service worker keepalive
webRequest - Network monitoring for capture/replay
webNavigation - Detect page loads, clear stale state
scripting - Content script injection for tech detection
management - Extension listing/reloading
<all_urls> - Required for automation on any site (your data NEVER leaves your machine in Free tier)

What We DON'T Collect:
Browsing history, page content, cookies, form inputs, passwords, personal info, analytics

Open Source:
• Full source: https://github.com/railsblueprint/blueprint-mcp
• Apache 2.0 license
• Audit the code yourself

PRICING

Free Tier - $0/month
• All 20+ browser automation tools
• Local WebSocket connection
• Unlimited usage
• Zero cloud dependencies
• Same machine only

PRO Tier - $5/month or $50/year ($1 for 14-day trial)
Everything in Free, plus:
• Cloud Relay (control from anywhere)
• Multiple browsers
• Multi-client access
• Auto-reconnection
• Priority support

Upgrade: https://blueprint-mcp.railsblueprint.com/pro

Enterprise - Custom pricing
Self-hosted relay, custom integrations, SLA guarantees. Contact: [email protected]

SUPPORT

Documentation: https://github.com/railsblueprint/blueprint-mcp
GitHub Issues: https://github.com/railsblueprint/blueprint-mcp/issues
Email: [email protected]

Troubleshooting:

Not connecting?
1. Check MCP server running: npx @railsblueprint/blueprint-mcp@latest --debug
2. Verify port 5555 available
3. Check extension popup shows "Connected"
4. Review debug logs in DevTools console

AI can't find tools?
1. Verify MCP server in AI client config
2. Restart AI client after config change
3. Run enable tool first to activate

Need help? Open GitHub issue with Edge/extension version, MCP server version, AI client used, and debug logs.

Blueprint MCP for Edge - Give your AI assistant the browser automation it deserves.

https://mcp-for-edge.railsblueprint.com
https://github.com/railsblueprint/blueprint-mcp