MCP Server Live — mcp.e2llm.com

Give AI Eyes.
And Hands.

One click captures live UI state. MCP Server lets AI see pages, click buttons, fill forms. Claude, GPT, Gemini, Codex — they act on what they actually see.

Install Extension Explore MCP → Prompts & Examples
The Problem
Your AI agent is blind
It guesses from screenshots. Drowns in raw HTML. Misses what users actually see.
📸

Screenshots

No element IDs. No computed styles. No z-index. No ARIA roles. Burns vision tokens on pixels.

No actionable selectors
📄

Raw HTML

2.3 MB of scripts, navs, widgets. Fills context before AI reads anything useful.

~600K tokens per page
🌳

Accessibility Trees

No computed styles, no positions, no z-index. What assistive tech sees ≠ what's rendered.

60% task accuracy
Two Products. One Perception Layer.
Choose how AI sees your browser
Manual capture for developers. MCP Server for autonomous AI agents. Same SiFR format, different workflows.
In stores now

E2LLM Extension

You pick an element — get a structured SiFR snapshot in clipboard or file. Paste into any LLM chat. Debug, review, audit.

  • Click any element → SiFR JSON in clipboard
  • Full-page capture with salience scoring
  • Save to disk — diff workflows, audit trails
  • Zero tokens until you paste — no overhead
  • 100% local — nothing leaves your browser
Chrome (92 users) · Firefox (13 users) · Arc · Brave · Edge
New — separate listing

E2LLM MCP

AI tools connect to your browser directly. They see pages, find elements, click buttons, fill forms, scroll. Eyes and hands.

  • AI sees live page via sifr_capture
  • AI clicks, fills, scrolls via actions
  • Works with Claude Code, ChatGPT, OpenAI Codex, Cursor, VS Code
  • 5–15KB per snapshot vs 200–400KB raw HTML
  • Tracks changes — modals, navigations, updates
  • SiFR rates element importance — AI ignores noise
MCP Server: mcp.e2llm.com · Authenticated
How MCP Works
AI connects to your browser in 4 steps
🤖
LLM Client
Claude · ChatGPT · Codex · Cursor
⇄
⚡
MCP Server
mcp.e2llm.com
⇄
🌐
Your Browser
Extension + SiFR engine
1

Install

Get E2LLM MCP from Chrome or Firefox store

2

Log in

Click extension icon, authenticate

3

Copy config

Extension shows MCP config — paste into AI tool

4

Done

AI is connected to your browser

What your AI can do now:
👁️
"Show me what's on the page"
→ structured SiFR snapshot, 5–15KB
🔍
"Find all buttons on this form"
→ list with IDs, labels, states, positions
👆
"Click Sign In"
→ extension clicks, AI sees result
✏️
"Fill email and password, submit"
→ extension types, submits, AI confirms
📜
"Scroll down and check the footer"
→ extension scrolls, captures new state
🔄
"What changed after I clicked?"
→ SiFR diff — modal opened, field updated
competitors vs e2llm
Playwright / Selenium
AI receives: raw HTML
200–400KB of noise
Scripts, nav menus, footers
No importance ranking
No spatial relationships

→ AI burns tokens on garbage
→ Misses what actually matters
E2LLM + SiFR
btn_003 "Submit" [clickable]
  salience: high
  position: (540, 320)
  stacked above: input_007
  occlusion: none

// 5–15KB. Importance-ranked.
// Typically 10–50x fewer tokens.
Use Cases
Your browser. Your sessions. AI does the rest.
No selectors, no scripts, no prompt engineering. Describe what you need in plain language.
🔐

Behind Logins & Paywalls

AI uses your browser with your sessions. Government portals, corporate intranets, private forums, paid subscriptions — if you can access it, AI can act in it.

🤷

"What Do They Want?"

Complex portal, unclear instructions, bureaucratic forms. Ask "what is this page asking me to do?" — AI explains and acts. Zero prompt engineering required.

🌏

Any Language, Any Script

DOM-based, not vision. CJK, RTL, Cyrillic work natively. Find a bus in rural Korea, read Chinese policy, navigate Arabic gov portals — zero language config.

🪟

Multi-Tab

AI sees all your tabs across browsers. Compare alternatives, cross-post between AI tools, research and leave tabs open for your review.

🚀

Ship It Yourself

Point AI at your staging URL. One prompt, full audit — bugs, UX issues, i18n, responsive breaks. You don't need a QA team.

♿

Accessibility

Every website becomes usable. AI acts on behalf of users who can't navigate complex UIs — motor, visual, cognitive, language barriers. Agency, not charity.

How It Compares
Right data beats more data
ScreenshotsRaw HTMLAXTreeE2LLM + SiFR
Element IDs✗✓partial✓ labeled
Computed Styles✗✗✗✓
Salience Scoring✗✗✗✓
Spatial Contextpixels✗✗✓ mapped
Token Costvision $$$~600K~15K~2–5K
Actions (click/fill)✗✗✗✓ MCP
Change Detection✗✗✗✓ diff
Pricing
Start free. Full features. Scale when you need to.
E2LLM Extension is always free. MCP Server tiers below — every tier includes actions.
Free
$0
forever
  • 100 operations / day
  • All features: capture, click, fill, scroll
  • SiFR format, full fidelity
  • Community support
Get Started
Pro
$19
per month
  • 250 operations / day
  • All features: capture, click, fill, scroll
  • Change detection & SiFR diffs
  • Priority server — no queue
  • Email support
Subscribe
Pro+
$50
per month
  • 750 operations / day
  • All features: capture, click, fill, scroll
  • Change detection & SiFR diffs
  • Priority server — no queue
  • 7-day diff history
  • Priority support
Subscribe
Enterprise
Custom
contact us
  • Unlimited operations
  • Self-hosted on your infrastructure — airgap ready
  • Persistent browser pool with identity (SSO/LDAP)
  • Capture-to-replay — LLM sessions become automation
  • Full audit trail & compliance controls
  • SLA & dedicated support
More Information
E2LLM Extension (manual capture) remains free at all tiers. Pricing applies to MCP Server only.
Data & Privacy
Different products. Different data models.
We believe you should know exactly what each product collects and why. No surprises.
E2LLM Extension

Minimal & Anonymous — your data stays local

  • SiFR generation runs entirely in your browser — DOM never leaves your machine
  • No account required. No login. No tracking.
  • We collect anonymous, aggregated usage analytics only: full-page vs partial capture, and capture frequency
  • We never see URLs, page content, or any captured data
  • Works fully offline — air-gap ready
Extension analytics help us improve capture quality. That's it. We don't know who you are, where you browse, or what you capture.
E2LLM MCP Server

Retained & Protected — responsible data handling

  • All session data (SiFR captures, actions) retained for 30 days — encrypted at rest, then purged
  • Required for safety monitoring, dispute resolution, and compliance
  • Encrypted before storage — infrastructure providers cannot read it
  • Access requires lawful request, user consent, or triggered safety review
  • 30-day retention is standard for AI platforms. Enterprise: self-hosted, you control retention.
We hold your data because we must — for safety and compliance. We don't access it unless there's cause. Encrypted, audited, purged after 30 days.
Get Started
Install. One click. AI sees.
Chrome & Arc Firefox Edge & Brave
Free. Privacy-first. Install in seconds.
Works with Claude · ChatGPT · Gemini · Grok · Llama · Cursor · Claude Code · OpenAI Codex · VS Code
v2.8.0 — Save captures to disk · Clipboard / File toggle · Persistent pipelines