Better Playwright MCP
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Better Playwright MCPGet the outline of the page"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Better Playwright MCP
Token-efficient Playwright MCP with getOutline and searchSnapshot - saves ~95% tokens vs full snapshots.
This is a fork of livoras/better-playwright-mcp with critical fixes applied.
Token Savings
Method | Lines | Characters | Savings |
Official Playwright MCP (full snapshot) | 673 | ~42,000 | - |
Better Playwright (getOutline) | 48 | ~2,100 | 95% |
Related MCP server: Enhanced Browser MCP Server
Fixes Applied
1. Snapshot Bug Fix
The original package's page._snapshotForAI() returns { full: "..." } object instead of string in newer Playwright versions, causing snapshot.split is not a function error.
// Before (broken):
const snapshot = await pageInfo.page._snapshotForAI();
// After (fixed):
const rawSnapshot = await pageInfo.page._snapshotForAI();
const snapshot = typeof rawSnapshot === 'string' ? rawSnapshot : rawSnapshot?.full ?? '';2. Missing MCP Server
The original npm package references dist/mcp-server.js which doesn't exist. This fork includes a proper MCP wrapper (index.mjs).
3. MCP SDK API Compatibility
The MCP SDK (v1.0+) changed from string-based to schema-based request handlers. This fork uses the new API:
// Old API (broken with MCP SDK v1.0+):
server.setRequestHandler('tools/list', async () => {...});
server.setRequestHandler('tools/call', async (req) => {...});
// New API (fixed):
import {
ListToolsRequestSchema,
CallToolRequestSchema,
} from '@modelcontextprotocol/sdk/types.js';
server.setRequestHandler(ListToolsRequestSchema, async () => {...});
server.setRequestHandler(CallToolRequestSchema, async (req) => {...});Requirements
Node.js >= 18.0.0
Installation
Global Installation (Recommended)
npm install -g github:ekuznetski/better-playwright-mcpPer-Project Installation
# npm
npm install github:ekuznetski/better-playwright-mcp
# pnpm
pnpm add github:ekuznetski/better-playwright-mcp
# yarn
yarn add github:ekuznetski/better-playwright-mcpSetup
Step 1: Start HTTP Server
The HTTP server must be running for the MCP to work.
# If installed globally
better-playwright-server
# If installed per-project
./node_modules/.bin/better-playwright-server
# Or with npx
npx better-playwright-serverBy default, the server runs on port 3102. Set PORT environment variable to change.
Step 2: Configure MCP
Claude Code (Project Config)
Create .mcp.json in your project root:
{
"mcpServers": {
"better-playwright": {
"command": "better-playwright-mcp",
"env": {
"BETTER_PLAYWRIGHT_URL": "http://localhost:3102"
}
}
}
}Claude Code (Global Config)
Add to ~/.claude.json:
{
"mcpServers": {
"better-playwright": {
"command": "better-playwright-mcp",
"env": {
"BETTER_PLAYWRIGHT_URL": "http://localhost:3102"
}
}
}
}Available Tools
Tool | Description |
| Create a new browser page and navigate to URL. Returns |
| Get compressed page structure (max ~200 lines). 95% token savings. |
| Search page content with regex. Returns only matching lines (max 100). |
| Click element by ref ID (e.g., "e5", "e12"). |
| Type text into element by ref ID. |
| Hover over element by ref ID. |
| Navigate to URL. |
| Take screenshot of page. |
| Press keyboard key (Enter, Tab, Escape, etc.). |
| Scroll to top of page. |
| Scroll to bottom of page. |
| Get browser console logs and JS errors. Supports |
| List all open browser pages. |
| Close browser page. |
Usage Guide for AI Agents
Recommended workflow
1. create_page(name: "app", url: "http://localhost:3000")
-> Returns pageId: "abc-123"
2. get_outline(pageId: "abc-123")
-> Returns compressed page structure with element refs like [ref=e5]
-> Use this to understand layout and find clickable elements
3. search_snapshot(pageId: "abc-123", pattern: "Submit|Login")
-> Returns only lines matching the pattern — much faster than get_outline for targeted search
4. click(pageId: "abc-123", ref: "e5")
-> Clicks element. Always get ref from get_outline or search_snapshot first.
5. get_console_messages(pageId: "abc-123", clear: true)
-> Check for JS errors or logs after interaction. Pass clear: true to flush buffer.When to use each tool
get_outline— first step to understand page structure, max ~200 linessearch_snapshot— when you know what text/element to find, much more token-efficientget_console_messages— after page load or interactions to catch JS errors, debug logsscreenshot— only when visual layout matters; costs tokens due to imagepress_key— for Enter (submit form), Escape (close modal), Tab (focus next field)
Debugging frontend issues
# After clicking a button or navigating, always check console:
get_console_messages(pageId: "abc-123")
# Expected output:
# [12:34:56.123] [LOG] Component mounted
# [12:34:56.200] [ERROR] Uncaught TypeError: Cannot read property 'id' of undefined
# To avoid accumulating old logs, clear after reading:
get_console_messages(pageId: "abc-123", clear: true)Finding elements
Refs in get_outline look like [ref=e5]. Use the ref directly in click, type_text, hover.
# Find a specific button:
search_snapshot(pageId: "abc-123", pattern: "Submit")
-> button "Submit" [ref=e42]
click(pageId: "abc-123", ref: "e42")Running HTTP Server as Background Service
macOS (launchd)
cat > ~/Library/LaunchAgents/com.better-playwright.plist << 'EOF'
<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0">
<dict>
<key>Label</key>
<string>com.better-playwright</string>
<key>ProgramArguments</key>
<array>
<string>/usr/local/bin/better-playwright-server</string>
</array>
<key>RunAtLoad</key>
<true/>
<key>KeepAlive</key>
<true/>
</dict>
</plist>
EOF
launchctl load ~/Library/LaunchAgents/com.better-playwright.plistLinux (systemd)
mkdir -p ~/.config/systemd/user
cat > ~/.config/systemd/user/better-playwright.service << 'EOF'
[Unit]
Description=Better Playwright HTTP Server
[Service]
ExecStart=/usr/local/bin/better-playwright-server
Restart=always
[Install]
WantedBy=default.target
EOF
systemctl --user enable better-playwright
systemctl --user start better-playwrightEnvironment Variables
Variable | Default | Description |
| 3102 | HTTP server port |
| URL for MCP to connect to HTTP server |
Troubleshooting
MCP not connecting / "Schema is missing a method literal"
If you see this error when Claude Code tries to connect:
Error: Schema is missing a method literalThis means the MCP SDK API changed. Ensure you have the latest version of this package:
npm update -g github:ekuznetski/better-playwright-mcpThe fix uses schema-based handlers instead of string-based ones (see "Fixes Applied" section above).
Server not starting
# Check if port is in use
lsof -i :3102
# Kill existing process
lsof -ti :3102 | xargs kill -9ripgrep binary missing
If you see /bin/sh: .../rg: No such file or directory:
cd node_modules/@vscode/ripgrep && npm run postinstallConnection refused
Make sure the HTTP server is running before using the MCP tools:
better-playwright-serverThe server should output: Better Playwright HTTP server running on port 3102
License
MIT
Credits
Original: livoras/better-playwright-mcp
Fork with fixes: ekuznetski/better-playwright-mcp
This server cannot be deployed
Maintenance
Related MCP Connectors
Live browser debugging for AI assistants — DOM, console, network via MCP.
Hosted browser for AI agents: screenshots, post-JS DOM, console, WCAG. No install, no API key.
Browser MCP for logged-in tasks. Uses your Chrome — credentials stay local. Zero-token replay.
Stealth web browser for agents: search, fetch, click, download and type in persistent MCP sessions.
Related MCP Servers
- FlicenseNot gradedqualityDmaintenanceAn advanced MCP server for browser automation using Puppeteer, specifically optimized for token efficiency through minimal data returns and progressive enhancement. It enables agents to navigate pages, capture LLM-optimized screenshots, extract structured content, and perform batch interactions.3-
- AlicenseAqualityDmaintenanceAn enhanced browser automation MCP server that solves token limit issues through intelligent semantic compression, enabling universal web automation with context-aware snapshot modes.121Apache 2.0
- AlicenseAqualityDmaintenanceA Playwright-powered MCP server for browser automation using ARIA snapshots and element refs, enabling LLMs to control Chrome/Edge without CSS selectors.425 npmMIT
- FlicenseNot gradedqualityDmaintenanceToken-efficient browser automation for AI agents, filtering the DOM to only interactive elements and grouping them by page section, reducing token usage by 3-13x compared to Playwright MCP.-