Oculo — open-source AI browser for MCP clients
Open Source · Free · MIT License · v0.4.4

The AI Browser That
Sees the Web

12 MCP tools. ~30 tokens per page description. Give AI eyes to see and hands to interact with any website.

Cursor:VSCode :: Oculo:Chrome

Electron 34 TypeScript React 19 MCP SDK Privacy-First

Built with

Electron TypeScript React Tailwind CSS MCP Protocol Whisper STT

Features

Everything AI needs to browse the web

Built from the ground up for AI agents. Twelve MCP tools, zero token waste, full security.

AI Chat Panel

Built-in AI sidebar with multi-provider support. Claude, OpenAI, Gemini, Grok, and OpenClaw. Chat about any page without leaving the browser.

12 MCP Tools

page act fill read run media shell tabs research preview translate lens — page descriptions in ~30 tokens. 99% fewer than DOM snapshots.

Smart Passkey & Auth

Credential vault backed by macOS Keychain with AES-256 encryption. AI says "login" — Oculo fills credentials securely. Passwords never touch the AI.

Media Generation

Generate images with Nano Banana and videos with Veo 3.1 directly from the browser. AI-powered creative tools in your workflow.

Privacy & Security

Four-tier permission gates. PII redaction on every MCP response. Anti-prompt-injection. Encrypted audit log. Touch ID for sensitive actions.

Voice Control

On-device Whisper speech-to-text. Talk to the AI or dictate form inputs. Also powers autonomous audio CAPTCHA solving — zero cloud calls.

New in v0.4.4

Zero-friction automation

Post a tweet in 1 tool call instead of 15. No permission prompts, no connection failures, no broken inputs.

Autonomous Mode

Auto-approve browser actions (click, type, fill, login). Payments, password changes, and shell commands still require explicit approval.

React/DraftJS Fill Fix

3-step fallback chain for contenteditable: clipboard paste, InputEvent, execCommand. Works on X.com, Notion, Slack, and other React apps.

2s Faster Per Call

Bridge connection caching with 30s TTL, /health endpoint, and auto-retry on disconnect. No more stale port file errors.

v0.4.0

8 new capabilities

Self-Healing Automation

Selector caching + DOM diffing. When a page changes, Oculo auto-recovers — no brittle selectors. 44%+ faster for repeated workflows.

Headless & Docker

Run without UI using --headless flag. Deploy as a Docker container for CI/CD pipelines and server-side automation.

Multi-Tab Parallel

New tabs tool to list, switch, and execute across multiple tabs simultaneously. True parallel browsing.

Proxy Support

HTTP/SOCKS proxy per session with encrypted credentials. Route traffic through any proxy for geo-targeting or privacy.

Session Recording

Record and replay MCP actions. Capture tool calls, results, and screenshots for debugging, auditing, or workflow templating.

Python SDK

pip install oculo — sync and async clients with auto-discovery. Control Oculo from any Python script.

Enhanced Stealth

19 anti-fingerprint patches — navigator, WebGL, canvas, WebRTC, audio, fonts, battery, and screen randomization. Browse undetected.

Cross-Platform

Native builds for macOS, Windows, and Linux. CI/CD pipeline auto-builds all platforms. AppImage + deb for Linux.

Open Source

Built in the open

Oculo is fully open-source under the MIT license. Inspect, modify, and redistribute freely.

MIT
License

No restrictions. Use it anywhere.

100%
Open Source

Every line of code on GitHub.

$0
Forever Free

No paid tiers. No subscriptions.

How It Works

Three steps to AI browsing

From install to AI-powered web interaction in under a minute.

01

Install Oculo

Download the macOS app or clone from GitHub. Launch it and Oculo starts as both a browser and an MCP server.

git clone https://github.com/xidik12/oculo.git
02

Connect Any MCP Client

Register Oculo as an MCP server in Claude Code, Cursor, Windsurf, or any MCP client. One command.

claude mcp add oculo ...
03

AI Browses For You

Ask Claude to navigate, fill forms, extract data, or run multi-step workflows. ~30 tokens per page.

"Search for flights to Tokyo"

MCP Integration

Connect in one command

Register Oculo as an MCP server and unlock 12 powerful tools for any AI client.

Terminal
# Register Oculo as an MCP server (works with any MCP client)
$ claude mcp add oculo -- node /path/to/oculo/bin/oculo-mcp.mjs
page See

Describe current page — headings, forms, buttons, links. Compact summaries in ~30 tokens, not DOM dumps.

act Do

Navigate, click, type, scroll, and 42+ actions. Intent-based with self-healing selectors.

fill Fill

Fill forms by label matching — name, email, country. Smart field detection with auto-submit option.

read Extract

Structured data extraction — search results, tables, lists, product details. Text or JSON output.

run Pipeline

Multi-step automation with conditionals — chain navigate, fill, click, wait, and read in one call.

media Generate

Images (Nano Banana, DALL-E 3) + video (Veo 3.1). Creative AI generation at your fingertips.

shell Execute

Run terminal commands (ls, npm, git, python). Non-interactive, capped output, timeout-protected.

tabs Manage

Multi-tab parallel execution. List, switch, open, and close tabs. Work across multiple pages at once.

research Research

Deep web research across tabs. Search, read multiple pages, and synthesize a markdown report with sources.

preview Preview

Pre-fetch URLs without navigating. Get a page description before committing to a full visit.

translate Translate

Page or text translation to any language. Browse the web without language barriers.

lens Vision

Visual page analysis via AI screenshot. Understand layouts, images, and visual content that text descriptions miss.

Example — AI agent using Oculo
# AI prompt (Claude Code, Cursor, Windsurf, etc.):
"Go to Hacker News and get the top 5 stories"

# Oculo handles it in ~3 tool calls:
act  → navigate to news.ycombinator.com        # 10 tokens
page → "Hacker News | 30 stories, nav, search"  # 30 tokens
read → extract top 5 with title, points, link  # 80 tokens

# Total: ~120 tokens (vs ~25,000 with DOM snapshots)
12
MCP Tools
~30
Tokens per page
99%
Less than DOM snapshots
5
AI Providers

Ready to give AI vision?

Oculo is free and available now for macOS, Windows, and Linux.

First launch on macOS?

Since Oculo isn't notarized by Apple yet, macOS will show a security warning. Here's how to open it:

  1. 1 Open the .dmg file and drag Oculo to Applications
  2. 2 Right-click (or Control-click) the Oculo app and select Open
  3. 3 Click Open in the dialog that appears

Alternatively: System Settings → Privacy & Security → scroll down and click "Open Anyway" next to Oculo. You only need to do this once.

Available on all platforms. Star on GitHub for updates.

Or clone from source

$ git clone https://github.com/xidik12/oculo.git && cd oculo && npm install && npm run dev

Privacy Policy

No telemetry. Oculo does not collect, transmit, or store any usage data, analytics, or personal information. There are no tracking pixels, session recordings, or third-party analytics services.

Local-only data. All browsing data (history, bookmarks, downloads, credentials) is stored locally on your device in ~/Library/Application Support/oculo/. No data is synced to any server.

Credential security. Passwords stored in the vault are encrypted using your operating system's keychain (macOS Keychain / Windows DPAPI) with AES-256 encryption. They are never exposed via IPC, MCP, or any API.

MCP privacy. All MCP responses are automatically processed through a PII redactor that strips credit card numbers, SSNs, JWTs, API keys, and other sensitive data before they reach any AI model.

External connections. This website loads Google Fonts (Inter). Oculo itself makes no external connections other than the websites you choose to visit.

Last updated: March 2, 2026 · Contact: GitHub Issues

FAQ

Frequently Asked Questions

Everything you need to know about the open-source AI browser for MCP.

What is Oculo and how is it different from other AI browser automation tools?
Oculo is a free, open-source AI browser built specifically for MCP (Model Context Protocol) clients. Unlike headless browser automation frameworks like Playwright or Puppeteer, Oculo is a real browser you use every day — with tabs, bookmarks, history, and a built-in AI chat panel. It has 12 MCP tools and describes pages in ~30 tokens compared to ~25,000 tokens for full DOM snapshots, making it dramatically more efficient for AI agents. Think of it as: Cursor is to VS Code what Oculo is to Chrome.
How do I connect Oculo to Claude Code, Cursor, or Windsurf?
One command: claude mcp add oculo -- node ~/path/to/oculo-mcp.mjs. Oculo runs as both your daily browser and an MCP server simultaneously. Any MCP-compatible client — Claude Code, Cursor, Windsurf, or custom agents — can connect and start browsing, clicking, filling forms, and extracting data through Oculo's 12 tools. No API keys required, no cloud dependency.
Is Oculo free and open source?
Yes — completely free, forever. Oculo is released under the MIT license with no paid tiers, subscriptions, usage limits, or telemetry. The source code is available on GitHub. You can inspect, modify, and redistribute it freely.
What platforms does Oculo support?
Oculo is available for macOS (Apple Silicon / arm64), Windows (x64), and Linux (AppImage / deb) as native installers. Built with Electron 34, TypeScript, and React 19 — the same stack used by Cursor, VS Code, and Slack.
What are the 12 MCP tools and how do they work?
Oculo provides 12 MCP tools for AI web automation: page (describe what's on screen), act (click, navigate, scroll, type), fill (fill forms by label), read (extract structured data), run (multi-step pipelines with conditionals), media (generate images and videos), shell (execute terminal commands), tabs (multi-tab parallel execution), research (deep web research), preview (pre-fetch URLs), translate (page translation), and lens (visual page analysis). Each tool is optimized for minimal token usage — a full page description costs ~30 tokens vs ~25,000 with traditional DOM approaches.
Is my data private? How does Oculo handle security?
Privacy is core to Oculo's design. All data stays on your device — there are no analytics, telemetry, or cloud services. Oculo uses a four-tier permission system (auto, notify, confirm, blocked) for MCP actions. PII is automatically redacted from every MCP response before it reaches any AI model. Passwords are stored in the OS keychain using AES-256 encryption and are never exposed via IPC or MCP. Anti-prompt-injection boundaries protect against malicious web content.
Can I use Oculo as my daily web browser?
Absolutely. Oculo is a full-featured Chromium-based browser with tabs, bookmarks, history, downloads, a built-in AI chat panel, reader mode, split view, and more. It's designed to be your primary browser — not just an automation tool. The AI capabilities are always available but never get in the way of normal browsing.