ChatGPT Escalation MCP Server

# Protocol Probe (MCP) This probe validates that the MCP server is discoverable and responds correctly to core protocol flows without relying on the CLI. What it does - Starts the server over stdio and performs: - initialize/connect - list tools and validate input schema for `escalate_to_expert` - call `list_projects` and parse JSON - call `escalate_to_expert` and assert a non-error, structured JSON response - Prints server logs from stderr for easy diagnosis - Exits with code 0 on success; non‑zero on failure Prerequisites - Windows with ChatGPT Desktop installed and signed in - A valid config with at least one project (see `chatgpt-escalation-mcp init`) - Python 3 with `pywinauto` and `pyperclip` installed - Install with: `npm run setup-python` Usage ```powershell # From repo root npm run probe # or node tools/mcp_protocol_probe.js ``` Expected output (abridged) ``` [INFO] [escalation-mcp:server] Starting ... [probe] Connected [probe] Tools: [ 'list_projects', 'escalate_to_expert' ] [probe] Projects OK [probe] Calling escalate_to_expert... [INFO] [escalation-mcp:chatgpt-desktop] Escalation completed successfully [probe] Escalation response OK [probe] SUCCESS ``` Troubleshooting - Probe fails at tools discovery: - Ensure Node 18+ and run `npm run build` if using from source. - `list_projects` returns non‑JSON: - Check your config file and run `chatgpt-escalation-mcp doctor`. - Escalation `isError=true` or `{ error: true }` JSON: - Confirm ChatGPT Desktop is open and responsive. - Verify projects are configured: `chatgpt-escalation-mcp init`. - Increase timeout in your config if responses are slow. - Ensure Python deps are present: `npm run setup-python`. Notes - The probe uses long timeouts (10m) for tool calls to accommodate slow UI operations. - Server logs print to stderr; redirect/stdout remains reserved for MCP protocol.