ChatMCP – AI 브라우저 채팅을 코딩 에이전트에 연결

Pull context from your AI browser tabs directly into Claude Code (and other AI coding tools) via MCP. Claude Code / Copilot CLI / Codex CLI ← stdio (spawns server as subprocess) OR Cursor / Antigravity / Copilot (VS Code) ← HTTP http://127.0.0.1:27183/mcp ↕ Node MCP Server ←── WebSocket on localhost:27182 ↕ Chrome Extension (background service worker) ↕ AI Chat Tabs (ChatGPT, Claude, Gemini, Grok, DeepSeek, Mistral, Perplexity) You need two things: a Chrome extension and a small server program. Here's exactly what to do. - Go to nodejs.org and download the LTS version - Run the installer — click Next through everything, keep all defaults - When it's done, open Terminal (Mac) or Command Prompt (Windows) and type: You should see something like node --version v20.x.x . If so, you're good. In Terminal / Command Prompt: npm install -g @anthropic-ai/claude-code Then run claude and follow the login prompt. Copy and paste this one line into your terminal: claude mcp add chatmcp -- npx -y chatmcp-server That's it. Claude Code will automatically start the server whenever you use it. - Download or clone this repository — click the green Code button on GitHub, then Download ZIP, and unzip it somewhere on your computer - Open Chrome and go to chrome://extensions in the address bar - In the top-right corner, turn on Developer mode - Click Load unpacked - Select the chrome-extension folder from what you just unzipped - You should see the ChatMCP extension appear with an OFF badge - Start Claude Code by typing claude in your terminal - The extension badge should turn ON (green) within a few seconds - Open any AI chat tab (ChatGPT, Gemini, etc.) in Chrome - In Claude Code, ask: "pull context from my browser tabs" Two modes: Tools that can spawn subprocesses use stdio (simpler — no background process needed). Tools that connect to a running server use HTTP mode — start the server once and leave it running. claude mcp add chatmcp -- npx -y chatmcp-server Copilot in VS Code (1.99+) spawns the server for you — no background process needed. Workspace — create .vscode/mcp.json in your project: { "servers": { "chatmcp": { "command": "npx", "args": ["-y", "chatmcp-server"] } } } Global (all projects) — open VS Code, run MCP: Open User Configuration from the Command Palette, and add: { "servers": { "chatmcp": { "command": "npx", "args": ["-y", "chatmcp-server"] } } } Then open Copilot Chat, switch to Agent mode, and the chatmcp tools appear automatically. (MCP tools are invisible in Ask / Edit mode.) Config lives at ~/.copilot/mcp-config.json (created automatically on first use). Add: { "mcpServers": { "chatmcp": { "type": "local", "command": "npx", "args": ["-y", "chatmcp-server"] } } } Or use the interactive command inside the CLI: /mcp add Config lives at ~/.codex/config.toml . Add a section: [mcp_servers.chatmcp] command = "npx" args = ["-y", "chatmcp-server"] Or scope it to a project by placing the same block in .codex/config.toml inside your repo. You can also use the CLI helper: codex mcp add Project-level — create .cursor/mcp.json in your project root: { "mcpServers": { "chatmcp": { "type": "stdio", "command": "npx", "args": ["-y", "chatmcp-server"] } } } Global — add the same block to ~/.cursor/mcp.json . Verify in Cursor Settings → Tools & MCP — you should see chatmcp with a green status indicator. Open an Agent session, click ⋯ → MCP Servers → Manage MCP Servers → View raw config, and add to mcp_config.json : { "mcpServers": { "chatmcp": { "command": "npx", "args": ["-y", "chatmcp-server"] } } } Save the file — Antigravity picks up the new server immediately. If your tool connects to MCP servers over HTTP/SSE rather than spawning subprocesses (OpenCode, Windsurf, custom agents, etc.), start the server in HTTP mode first: npx -y chatmcp-server --http The server runs persistently on http://127.0.0.1:27183/mcp and accepts multiple simultaneous clients. Point your tool at that URL. Keep the terminal open (or add it to your system startup / launchd / systemd ). git clone https://github.com/indiantinker/chatMCP.git cd chatMCP/mcp-server npm install && npm run build # stdio (Claude Code, Copilot, Codex, Cursor, Antigravity) node /path/to/chatMCP/mcp-server/dist/index.js # HTTP (OpenCode, Windsurf, or any HTTP MCP client) node /path/to/chatMCP/mcp-server/dist/index.js --http For Claude Code with a local build: claude mcp add chatmcp -- node /path/to/chatMCP/mcp-server/dist/index.js # See what AI chats are open list_ai_tabs # Pull the full conversation (auto-selects active AI tab) get_chat_context # Pull from a specific tab get_chat_context tabId=12345 # Get a summary instead of the full transcript get_chat_context summaryOnly=true # Read any tab (docs, GitHub issues, articles) get_page_content # Check extension status extension_status Or just ask naturally: "Pull context from my ChatGPT tab and continue building on that idea." | Platform | URL | |---|---| | ChatGPT | chat.openai.com / chatgpt.com | | Claude | claude.ai | | Gemini | gemini.google.com | | Grok | grok.com | | DeepSeek | chat.deepseek.com | | Mistral | chat.mistral.ai | | Perplexity | perplexity.ai | All data stays on your machine. The MCP server and Chrome extension communicate over localhost only — nothing is sent to any external server. Chat content is passed directly to your local Claude Code session. - The MCP server starts automatically when Claude Code launches if registered (stdio mode). - In HTTP mode, start the server manually and keep it running. - The extension auto-reconnects every 3 seconds if the server restarts. - Chat extraction uses DOM scraping — may need updates when AI platforms change their markup. MIT — see LICENSE. If this saves you time, buy me a coffee. ☕