contextbridge

Shared context layer for AI-native developers. Auto-capture conversations from ChatGPT and Gemini, expose them to Cursor and Claude Code via MCP, and seamlessly inject relevant context back into browser tools.

Problem

Developers live in fragmented AI tool silos:

• ChatGPT conversation → manual copy-paste to Cursor
• Gemini image → manual screenshot or download
• Cursor decision → no way to send back to ChatGPT

contextbridge is the missing glue.

How it works

┌──────────┐ capture  ┌─────────────┐ inject  ┌─────────────┐
│ ChatGPT  │ ────────►│ Context     │ ◄──────┤ Cursor      │
│ Gemini   │          │ Store       │        │ Claude Code │
└──────────┘          └─────────────┘        └─────────────┘

1. Capture: Chrome extension watches ChatGPT/Gemini conversations
2. Store: Conversations saved as markdown files + indexed
3. MCP Expose: Cursor/Claude Code read via MCP tools
4. Inject: Confirmation toast offers to inject context back into tools

Quick Start

Requirements

• Node.js 18+
• npm
• Chrome browser
• Cursor or Claude Code (optional, for MCP integration)

Setup (5 minutes)

# 1. Install dependencies
npm install
cd extension && npm install && cd ..

# 2. Build bridge + extension
npm run build
cd extension && npm run build && cd ..

# 3. Load extension into Chrome
#    - chrome://extensions/
#    - Enable Developer mode
#    - Load unpacked → select contextbridge/extension/dist/

# 4. Install MCP for Cursor/Claude Code (optional)
bash scripts/install-mcp.sh

# 5. Start the bridge
npm run dev

The bridge runs on http://127.0.0.1:3131.

Your First Test

1. Open https://chat.openai.com
2. Have a conversation with ChatGPT
3. Watch ~/.contextbridge/sessions/ for new .md files appearing in real-time
4. Open https://gemini.google.com
5. A toast should appear: "📋 New context from chatgpt"
6. Click "Yes, inject" to insert the ChatGPT discussion into your Gemini prompt

Testing Week 1 (manual capture)

# Terminal 1: start bridge
pnpm dev

# Terminal 2: capture a test event
curl -X POST http://127.0.0.1:3131/capture \
  -H 'Content-Type: application/json' \
  -d '{
    "source": "manual",
    "title": "JWT strategy decision",
    "url": "",
    "content": "Use short-lived access tokens + httpOnly refresh tokens for security."
  }'

# Verify the file was created
ls ~/.contextbridge/sessions/
cat ~/.contextbridge/index.json

Then in Cursor or Claude Code:

You: "Use get_recent_context to tell me what's been captured"

Cursor/Claude Code will call the MCP tool and return your captured context.

Architecture

Component	Purpose
Bridge (Fastify + MCP)	Captures events via REST, exposes MCP tools to IDEs
Markdown Store	All context stored in ~/.contextbridge/sessions/*.md (git-friendly)
Chrome Extension	(Week 2+) Captures DOM from ChatGPT/Gemini
Confirmation Toast	(Week 4+) Offers to inject context back into tools

Directory structure

~/.contextbridge/
  sessions/               ← one markdown file per captured context
    2026-04-11_chatgpt_*.md
    2026-04-11_gemini_*.md
  artifacts/              ← images from Gemini
  index.json              ← index of all sessions

Development

# Type check
pnpm typecheck

# Lint
pnpm lint

# Watch mode (rebuilds on change)
pnpm --filter bridge dev

Status

v0.1.0 - COMPLETE ✅

All core features implemented and tested:

• Week 1 ✓ Bridge + markdown store
• Week 2 ✓ Chrome extension: ChatGPT + Gemini capture
• Week 3 ✓ Image artifacts + linking
• Week 4 ✓ Bidirectional injection with toast UI
• Week 5 ✓ MCP integration + Cursor rules

See detailed docs:

• WEEK2_SETUP.md — Extension setup
• WEEK3_ARTIFACTS.md — Image capture
• WEEK4_BIDIRECTIONAL.md — Injection flow
• WEEK5_MCP_INTEGRATION.md — MCP + Cursor rules

MCP Tools (exposed to Cursor/Claude Code)

Tool	Purpose
get_recent_context(limit, source?)	Get recent captured contexts
search_context(q, limit)	Search by title/tags
get_session(id)	Get full body of a session
save_context(title, content, tags?)	Write a decision to store
list_artifacts(sessionId?)	List captured images

Usage in Cursor Rules

Add to .cursor/rules or your workspace config:

At the start of each task, call `get_recent_context` to check for relevant prior discussions or decisions from ChatGPT/Gemini.

License

MIT

Contributing

Contributions welcome! Please open an issue or PR.