home / mcp / chatgpt escalation mcp server

ChatGPT Escalation MCP Server

MCP server to escalate hard questions to the ChatGPT Desktop app on Windows via automated UI interactions.

Installation

Add the following to your MCP client configuration file.

Configuration

{
  "mcpServers": {
    "dazlarus-chatgpt-escalation-mcp": {
      "command": "node",
      "args": [
        "N://AI Projects//chatgpt-escalation-mcp//dist//src//server.js"
      ],
      "env": {
        "PATH": "PLACEHOLDER"
      }
    }
  }
}

You deploy an MCP server that lets autonomous coding agents escalate hard questions to the ChatGPT Desktop app on Windows. It automates the ChatGPT workflow inside the desktop app, sending prompts, waiting for responses, and returning results to your agent so it can continue work without manual intervention. This approach keeps automation ToS-compliant by operating only within your local ChatGPT Desktop instance and avoiding web UIs.

How to use

Use this MCP server with your MCP client to empower agents like Copilot or Roo to escalate difficult questions to ChatGPT for expert guidance. You configure one or more projects that map to specific ChatGPT conversations, then issue escalation requests from your agent with a project identifier and the question. Each escalation runs in a fresh ChatGPT session, returns the answer, and lets the agent proceed.

How to install

Prerequisites you need before installation include Node.js and Python, plus a Windows environment where the ChatGPT Desktop app is installed.

Step 1: Install required software

- Node.js 18 or newer

- Python 3.10 or newer

- Windows ChatGPT Desktop (Microsoft Store version) installed and signed in

Step 2: Install the MCP server from npm

# Install globally
npm install -g chatgpt-escalation-mcp

Step 3: Install Python dependencies used by the automation driver

pip install pywinauto pyperclip paddleocr paddlepaddle

Step 4: Run the setup wizard for initial configuration

chatgpt-escalation-mcp init

Step 5: (Optional) Build from source or install via release if you prefer that path

# From GitHub Release path
# Follow release-specific instructions to install and build

Additional sections

Configuration details describe how you organize projects, conversations, and timeouts. You place the main config file at the suggested path and then reference it from your MCP client.

Security and usage notes emphasize that the automation strictly targets the ChatGPT Desktop window, requires you to avoid interacting with the ChatGPT window during escalations, and supports a single escalation at a time.

Troubleshooting tips cover common issues like window detection, conversation lookup, and timeouts. Most problems resolve with retries, correct project/conversation naming, or updating the PaddleOCR model.

The server includes automatic retries and structured observability, assigning a unique run_id to each escalation for correlation and debugging.

Available tools

escalate_to_expert

Send a question to ChatGPT via the desktop app to obtain expert guidance for complex coding problems.

list_projects

Discover available project IDs and their ChatGPT conversation mappings from the configuration.