Conduit — The AI Data Bridge

Conduit — The AI Data Bridge Conduit organizes the massive streams of data flowing through your browser into structured, actionable insights — serving them instantly to your AI agents, automation scripts, and local tools. Whether you're automating complex browsing tasks programmatically, pulling raw HTML ASTs, or giving LLMs direct visibility into your live browsing window — Conduit turns your browser into a personal, sovereign data engine. 🔥 KEY FEATURES 🤖 MCP Server — Native AI Agent Integration Conduit inherently speaks MCP (the Model Context Protocol). Give Claude Desktop, Cursor, Windsurf, or your own scripts absolute real-time awareness of what your browser sees: • "Summarize the articles in my open tabs" • "Read these page elements and execute actions" • "Audit this page for accessibility issues" 24 purpose-built MCP tools give your agents total browser sovereignty: 🔍 Page Analysis & Auditing • `read_page` & `page_text`: Read visible DOM and extract text up to 15KB. • `read_raw_dom`: Pierce Shadow DOM boundaries and extract raw HTML AST directly. • `audit_a11y`: Run deep WCAG accessibility audits using the injected axe-core engine. • `page_meta` & `page_images`: Pull OpenGraph tags, favicons, and collect image URLs. • `screenshot`: Capture high-quality PNG/JPEG snapshots of active tabs. • `get_selection`: Read user-highlighted text intuitively. 🖱️ Remote Browser Control • `browser_action` & `execute_js`: Click elements, inject text, scroll natively, and execute arbitrary JS (with synchronous evaluated returns sent directly to your agent). • `keypress`: Simulate keyboard actions (Enter, Esc, Tab, Arrows). • `list_tabs`, `create_tab`, `close_tab`, `navigate_tab`: Total tab lifecycle control. 💾 Captured Data & History • `browse_data` & `drain_data`: Access and stream structured API captures. • `get_cookies`: Extract httpOnly cookies for agentic authentication workflows. • `bookmarks` & `browser_history`: Search native bookmarks and history seamlessly. ⚡ Full Local REST API Every capability is accessible via clean REST endpoints on localhost:7777: • GET /tabs — list open browser tabs • POST /action — click, type, scroll, and interact with pages • POST /screenshot — capture visual snapshots • GET /browser-history — search native browser history 🔐 YOUR DATA STAYS WITH YOU Zero cloud. Zero telemetry. The extension communicates exclusively with a local Node.js host server on 127.0.0.1. Nothing leaves your machine. You own your data completely. 📦 EDITIONS & TIERS All users activate a comprehensive 14-day Pro Trial on first install — full access to every REST endpoint, automation tool, and MCP bridge. After trial: core data capture, history, and read-only MCP tools remain free. Pro unlocks autonomous actions, tab control, cookies, and the WebSocket firehose. 🏗️ ARCHITECTURE • Extension: Lightweight MV3 service worker — passive data interception + recipe-driven extraction • Host: Local Node.js server — REST API (:7777), WebSocket stream (:7778), MCP Streamable HTTP/SSE (:7779) • Storage: Append-only JSONL data lake — persistent, portable, zero-dependency Designed by Sophia Labs. https://sophialabs.gantlett.io/conduit --- CHANGELOG (v2.1.173 - v2.1.175) • v2.1.175 (UI Modernization & Host Parity): Streamlined popup interface focused exclusively on AI Data Bridge endpoints and live host health. Synchronized internal host versions to eliminate false update alerts. Fixed Windows PE metadata patching sequence for standalone Single Executable Application (SEA) binaries. • v2.1.174 (Host Lifecycle & Process Isolation): Removed automatic connectNativeHost ignition on service worker boot to prevent duplicate background processes during remote port forwarding. Added EPIPE/ECONNRESET stream guards for disconnected stdio pipes. • v2.1.173 (Features & Refactors): Replaced external SDK with a lightweight, fetch-based SSE JSON-RPC parser in the Conduit client for robustness and speed. Migrated specialized endpoints into a unified /action endpoint pattern.