Setup Guide

From zero to first MCP capture in 5 minutes

Connect your AI agent to your browser. Everything you need — install, authenticate, configure, capture.

Install the extension

Install E2LLM from your browser's extension store:

Tip: After installation, you'll see the E2LLM icon in your browser toolbar. Click it to open the extension popup.

Create an account & sign in

Click the E2LLM extension icon and sign in with Google, Microsoft, GitHub, or email. Your account starts on the Free tier — all features included. See current limits.

The extension popup will show your account status and a green "Connected" indicator once authenticated.

Add MCP config to your AI tool

Copy the MCP configuration from your dashboard and paste it into your AI tool's settings.

Claude Code:

// Add to Claude Code MCP settings:

"e2llm": {
  "type": "url",
  "url": "https://mcp.e2llm.com/mcp"
}

OpenAI Codex:

# Run in terminal:
codex mcp add e2llm --url https://mcp.e2llm.com/mcp

Cursor / VS Code:

// Add to MCP settings:

"e2llm": {
  "url": "https://mcp.e2llm.com/mcp"
}

Your first capture

Open any website in your browser. Then ask your AI agent:

"Capture the current page and tell me what you see."

The agent will call sifr_capture, receive a structured SiFR snapshot of the page, and describe what's on screen — elements, buttons, forms, content. All in 2-5K tokens instead of 200-400K raw HTML.

That's it. Your AI agent can now see and act in your browser. Try: "Click the search button", "Fill in the email field", "Scroll down and read the footer."

Troubleshooting

Extension shows "Disconnected": Refresh the page (F5), then try again. If still disconnected, close and reopen Chrome/Firefox.

"No response from content script": The page's content script may have been unloaded. Refresh the target page and retry.

AI tool can't connect: Make sure the extension is signed in with the same account. Check that the MCP config URL is exactly https://mcp.e2llm.com/mcp.

Known limitation: Some custom date pickers, rich text editors (TipTap/ProseMirror), and heavily dynamic widgets may require multiple capture-act cycles. The agent will adapt automatically.