About Puppeteer MCP
Puppeteer MCP is a browser automation MCP server for developers and coding agents. Its single `execute` tool runs custom JavaScript with access to the Puppeteer Browser API, including across persistent sessions.
Key features
- 01Run JavaScript through one `execute` tool
- 02Access the Puppeteer Browser API
- 03Reuse browser sessions across calls
- 04Configure headless mode and viewport
- 05Pass custom flags to Chrome
Use cases
- 01Automate multi-step browser workflows
- 02Scrape content from web pages
- 03Test web applications with Puppeteer
MCP support
It exposes an `execute` tool to MCP clients, running supplied JavaScript in a Node.js VM with a `browser` global. The VM is not a security sandbox; the project is intended for trusted local development.
Configuration
Local server config is the same for every client. Copy it into your client’s MCP config file.
Tools
1Saved tool list · updated 3w ago
| Parameter | Type | Required | Description |
|---|---|---|---|
| code | string | Yes | JavaScript code to execute. Return JSON-serializable data |
| persistent | boolean | No | Reuse browser profile across calls; false creates isolated profile per call |
Details
From the project READMEpptr-mcp
MCP server for browser automation via Puppeteer. Unlike other browser MCPs that expose fixed tools (navigate, click, screenshot), this server executes arbitrary JavaScript code with direct access to the Puppeteer Browser instance.
Key Difference
Most browser MCP servers provide a limited set of predefined actions. This approach requires multiple round-trips for complex workflows and can't handle edge cases.
pptr-mcp takes a different approach: it exposes a single execute tool that runs your JavaScript code in a Node.js VM with a browser global. You write Puppeteer code directly, getting full API access in one call.
Traditional MCP (5 round-trips) pptr-mcp (1 round-trip)
================================ ================================
Agent Server Agent Server
| | | |
|-- navigate --->| |-- execute ---->|
|<-- ok ---------| | |
| | | +------------------------+
|-- waitFor ---->| | | const page = await |
|<-- ok ---------| | | browser.newPage(); |
| | | | await page.goto(url); |
|-- click ------>| | | await page.click(s); |
|<-- ok ---------| | | await page.type(i, t); |
| | | | return await |
|-- type ------->| | | page.screenshot(); |
|<-- ok ---------| | +------------------------+
| | | |
|-- screenshot ->| |<-- result -----|
|<-- image ------| | |
| | | |
Requirements
- Node.js >= 20
Installation
npm install -g pptr-mcp
MCP Configuration
Add to your MCP client config (e.g., Claude Desktop):
{
"mcpServers": {
"puppeteer": {
"command": "npx",
"args": ["pptr-mcp"]
}
}
}
With CLI options:
{
"mcpServers": {
"puppeteer": {
"command": "npx",
"args": ["pptr-mcp", "--no-headless", "--viewport=1080p"]
}
}
}
With custom Chrome flags (after --):
{
"mcpServers": {
"puppeteer": {
"command": "npx",
"args": [
"pptr-mcp",
"--viewport=1280x720",
"--",
"--proxy-server=http://proxy:8080"
]
}
}
}
CLI Options
| Option | Description |
|---|---|
--no-headless | Run with visible browser window |
--viewport=VALUE | Set viewport size (e.g., 1920x1080 or 1080p) |
--help, -h | Show help |
-- [args] | Pass remaining args to Chrome |
Unknown options before -- are also passed to Chrome.
Claude Code Plugin
Install as a Claude Code plugin:
/plugin marketplace add iatsiuk/pptr-mcp
/plugin install pptr-mcp@pptr-mcp
Environment Variables
| Variable | Description |
|---|---|
CHROME_PATH | Path to Chrome executable |
PUPPETEER_EXECUTABLE_PATH | Alternative to CHROME_PATH |
PUPPETEER_CACHE_DIR | Browser download cache directory |
PPTR_MCP_TIMEOUT | Execution timeout in ms (default: 30000) |
Tool: execute
Executes JavaScript code with access to Puppeteer browser.
Parameters
| Name | Type | Default | Description |
|---|---|---|---|
code | string | required | JavaScript code to execute |
persistent | boolean | true | Reuse browser session across calls |
Recipes
Disable headless mode
Show browser window during execution:
{
"mcpServers": {
"puppeteer": {
"command": "npx",
"args": ["pptr-mcp", "--no-headless"]
}
}
}
Custom Chrome profile directory
Use your own Chrome profile with saved logins and cookies:
{
"mcpServers": {
"puppeteer": {
"command": "npx",
"args": ["pptr-mcp", "--", "--user-data-dir=/path/to/profile"]
}
}
}
Use system Chrome instead of downloaded
By default, pptr-mcp downloads Chrome for Testing - a version optimized and tested for the bundled Puppeteer. To use your system Chrome instead:
{
"mcpServers": {
"puppeteer": {
"command": "npx",
"args": ["pptr-mcp"],
"env": {
"CHROME_PATH": "/path/to/chrome"
}
}
}
}
Security
This server is designed for trusted local development with LLM assistants (Claude Code, Cursor, etc.). The executing code comes from the LLM at your request.
What this means
- Not a sandbox: The Node.js VM isolates code for convenience, not security. It is not designed to run untrusted code.
- Full browser control: Executed code can navigate to any URL, read page content, take screenshots, and interact with web applications.
- Chrome runs without sandbox:
--no-sandboxflag is used for Docker/container compatibility. - Persistent sessions: With
persistent: true(default), cookies and browser state are preserved across calls. Usepersistent: falsefor isolation.
Not designed for
- Multi-tenant or shared server deployments
- Executing untrusted code from external sources
- Building web services that accept arbitrary user input
License
Comments
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.