MCP servers for a browser automation / testing agent
The MCP server bundle to build an AI that navigates and tests web apps in a real browser.
3 servers · ~28 tools · Last updated June 17, 2026
TL;DR: A browser agent needs a real engine to navigate, interact and capture state. This bundle offers a local option and a hosted option so you can start free and scale to managed, anti-bot-resistant browsers. Heads-up: browser servers are tool-heavy — check the combined tool count.
Bottom line: start with Playwright MCP Server and add the rest as your needs grow. All 3 install together via the merged config below (~28 tools total).
Tool budget: this stack exposes about 28 tools. That's within Cursor's practical ~40-tool ceiling, so all servers can stay enabled together. Check your own config →
What's in the stack
One-click config
All 3 servers merged into a single block — pick your client and paste.
{
"mcpServers": {
"playwright-mcp-server": {
"command": "npx",
"args": [
"-y",
"@playwright/mcp@latest"
]
},
"browserbase-mcp-stagehand": {
"command": "npx",
"args": [
"-y",
"@browserbasehq/mcp"
],
"env": {
"GEMINI_API_KEY": "<your-gemini-api-key>",
"BROWSERBASE_API_KEY": "<your-browserbase-api-key>",
"BROWSERBASE_PROJECT_ID": "<your-browserbase-project-id>"
}
},
"puppeteer-mcp-server-community": {
"command": "npx",
"args": [
"-y",
"puppeteer-mcp-server"
]
}
}
}Paste into ~/Library/Application Support/Claude/claude_desktop_config.json and fully restart Claude Desktop. Replace any placeholder keys/tokens with your own.
Capabilities this stack covers
FAQ
Local or hosted browser automation?
Local (Playwright/Puppeteer) is free and private; hosted (Browserbase) handles scale, proxies and anti-bot. Pick per workload.
Won't three browser servers blow the tool limit?
Likely — you usually pick one. Use the Config Doctor to check the combined tool count against Cursor's ceiling.