Skip to main content

Introduction

You're already logged into dozens of web apps in Chrome. Why should your AI need a separate set of credentials?

Every web app has internal APIs — the same endpoints its own frontend calls. I asked Claude to reverse-engineer them and expose them as MCP tools your AI agent can call directly. No API keys. No OAuth. No service accounts. If you can see it in a tab, your agent can use it.

No screenshots. No DOM scraping. No pixel-guessing. Your AI calls discord_send_message and it hits the same backend Discord's web app uses — running in your browser, using your existing session. It's fast, cheap on tokens (tokens are money), and the knowledge gets packaged into a reusable plugin anyone can install.

Works with Claude Code, Cursor, Windsurf, OpenCode, and any MCP client that supports Streamable HTTP.

Who Is This For?

Want to use OpenTabs? Start with the Quick Start — you'll go from install to a working tool call in under 5 minutes.

Want to build a plugin? The Plugin Development guide walks you through creating, building, and publishing a plugin from scratch.

Want to contribute to the platform? The Development Setup guide gets you running the full monorepo locally.

How It Works

Those internal APIs need to run inside your browser — that's where your sessions live. So the architecture has two halves: an MCP server that your AI talks to, and a Chrome extension that actually executes the calls in the right tab.

  1. Your AI sends a tool call — send a message, create an issue, query metrics, whatever you need. It's just a normal MCP tool call.

  2. OpenTabs routes it to the right tab — the MCP server finds the matching browser tab and dispatches the call through the Chrome extension.

  3. It runs on the real web app — the plugin adapter executes the action in the page using your logged-in session. Same API call the web app's own frontend makes. Results flow back to the agent.

What's Included

There are 100+ plugins covering ~2,000 tools across Slack, Discord, GitHub, Jira, Notion, Figma, AWS, Stripe, Robinhood, DoorDash, Airbnb, Netflix — and a bunch more. Each one talks to the real web app through your authenticated session. I built every one of them with strict end-to-end testing as a must-pass requirement in the build-plugin skill. The ones I use daily — Slack, GitHub, Discord, Todoist, Robinhood — I've personally verified and they work. For the rest, I relied fully on Claude to do the end-to-end testing. If something's broken, point your AI at it and open a PR. My AI will review what your AI wrote, and we'll merge it together.

Plus built-in browser tools that work on any tab — screenshots, clicking, typing, scrolling, network capture, cookies, DOM inspection, and more. No plugin needed for those.

Key Features

  • Your sessions, not your keys — uses your existing browser sessions. No API keys to manage, no OAuth flows to set up, no service accounts to provision. You're already logged in.
  • Real API calls, not browser automation — plugins call the web app's internal APIs directly. Structured, typed, deterministic. No DOM scraping, no pixel-guessing, no "click the third button from the left."
  • Plugins are just npm packages — anyone can create and publish OpenTabs plugins. Install with a single command and they're auto-discovered by the server.
  • Built-in browser tools — tab management, screenshots, DOM interaction, network capture, cookies, console logs — available for every tab, no plugin required.
  • Hot reload — the MCP server detects plugin changes automatically. Build, and it's live.

A Note on Trust

My honest recommendation: build your own plugins instead of installing third-party ones. Plugins run in your browser with your logged-in sessions — you should know what the code does. Point your AI agent at any website and it'll build a plugin for you in minutes. Your agent wrote it, you can read every line, and it runs in your browser with your session. That's a pretty good trust model.

If you do install a plugin from someone else, the system has your back: every plugin starts disabled, and when you enable one, your AI reviews the source code first. But nothing beats code you wrote (or watched your AI write) yourself.

Next Steps

Ready to get started? Head to the Quick Start — it takes less than 5 minutes.

Last Updated: 10 Mar, 2026