Tools_ForSharing

Chrome AI Assistant side panel + local knowledge MCP server — a browser extension that captures page context and fan-outs queries to multiple local AI CLIs in parallel, then synthesizes a unified answer.

Highlights

• Chrome Side Panel Extension (Manifest V3): Captures current page context with a 3-layer intelligent fallback (selection → Readability → raw DOM)
• Multi-Model Fan-Out: Sends the same query to multiple AI CLIs (Claude, Codex, Gemini) in parallel via a local bridge server
• Streaming Synthesis: Primary model synthesizes results from all responders with Markdown rendering (marked.js + highlight.js)
• Bridge Server: Node.js HTTP server (port 18080) bridges Chrome extension requests to local MCP/CLI tools
• Companion Knowledge MCP: Lightweight Python MCP server for local knowledge retrieval (FTS5 + vector search)

Tech Stack

Component	Technology
Extension	JavaScript, Chrome Manifest V3, Side Panel API
Bridge Server	Node.js, HTTP, CORS
Knowledge Server	Python, MCP SDK, SQLite FTS5, sentence-transformers
Rendering	marked.js, highlight.js
AI Integration	ai-cli-mcp (MCP protocol)

How It Works

┌─────────────────────┐
│  Chrome Browser Tab  │
│  (page context)      │
└─────────┬───────────┘
          │ 3-layer context extraction
┌─────────▼───────────┐
│  Side Panel UI       │
│  (query input +      │
│   streaming output)  │
└─────────┬───────────┘
          │ HTTP POST
┌─────────▼───────────┐
│  Bridge Server       │
│  (localhost:18080)    │
├─────────┬───────┬────┤
│ Claude  │ Codex │ …  │  ← parallel fan-out via ai-cli
└─────────┴───────┴────┘
          │ streamed results
┌─────────▼───────────┐
│  Synthesis + Render  │
│  (Markdown → HTML)   │
└─────────────────────┘

Setup

1. Load the extension: Chrome → chrome://extensions/ → Load unpacked → select chrome-plugin/
2. Start the bridge server:

   cd chrome-plugin/bridge-server
   node server.js
   # or use Start_Server.bat on Windows

3. Ensure ai-cli-mcp is installed and configured with at least one AI CLI tool

Design Decisions

• Bridge server architecture: Chrome extensions cannot spawn local processes directly. The bridge server acts as a gateway between the browser sandbox and local AI CLI tools.
• 3-layer context fallback: User selection → Readability-extracted article → raw DOM truncation. Ensures useful context regardless of page structure.
• Parallel fan-out: All configured AI models receive the query simultaneously. The primary model sees other models' responses for synthesis, producing a multi-perspective answer.