Provides tools for using a ChatGPT (OpenAI) browser session to queue and run coding, review, analysis, or image tasks. Supports submitting tasks, polling status and event history, cancelling work, retrieving response artifacts, and inspecting metrics and errors. It uses the user's ChatGPT account/session rather than the OpenAI API.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@Wrapper generate a TypeScript LRU cache with a usage example and poll until it finishes"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
I ran out of Codex usage, so I literally wrapped ChatGPT.
Connect your own ChatGPT browser session to coding agents using MCP. Wrapper runs locally, sends tasks through a real Chromium browser, and returns text, code, and image artifacts to your client.
Watch the 40-second demo · Install · npm · Documentation
The demo is an edited product presentation using captured results; it isn't an uncut recording. No API key is required, but you must sign in with your own ChatGPT account.
Requirements: Node.js 24+, npm, a desktop environment, and a ChatGPT account. The full setup and live coding workflow have been verified on macOS; Windows and Linux still need broader testing.
1. Install and connect to Codex
npx -y @rnaefe/chatgpt-wrapper@0.6.0The installer downloads Chromium, configures the Wrapper MCP server in Codex, and backs up any existing configuration before making changes. No clone or manual MCP config editing needed.
2. Sign in once
Complete the ChatGPT sign-in in the browser window. Wrapper keeps the session in a separate local profile. Browser verification, if requested, must be completed manually.
3. Restart Codex and try a task
Use the Wrapper MCP server to generate a TypeScript LRU cache with a short usage example. Check the task until it completes, then show me the result.
Your coding agent sends the task via MCP; Wrapper returns the result for review. It does not run generated code or change your project files.
Using a different client?
npx -y @rnaefe/chatgpt-wrapper@0.6.0 setup --client cursor
npx -y @rnaefe/chatgpt-wrapper@0.6.0 setup --client claude-desktopThese commands configure the selected MCP client; support for its configuration format doesn't imply every OS and browser combination has been tested. See installation and custom paths.
Related MCP server: AKIMCP
MCP interface: submit coding or analysis tasks, inspect their status, cancel work, and retrieve results.
Durable task queue: SQLite stores tasks and artifacts; interrupted browser work isn't silently submitted twice.
Browser session management: Playwright opens your saved ChatGPT session for work and closes the window afterward.
Developer tooling: a CLI for setup, reconnecting, updates and uninstall, plus local task diagnostics.
flowchart LR
A[Coding agent] --> B[MCP server]
B --> C[Local SQLite queue]
C --> D[Playwright worker]
D --> E[Your ChatGPT session]
E --> F[Result and artifacts]
F --> AWrapper is a task bridge, not another autonomous coding agent. Your MCP client handles planning, source files, applying changes, and reviewing generated output.
Version 0.6.0 includes code generation, code review, reasoning, image generation, image analysis and image editing, plus wait_task, get_status and chat-ready image exports. Image analysis and editing require input images; live provider access remains dependent on the account and browser UI. See MCP usage for task schemas and releases for shipped versions.
Version 0.6.0 also includes: selected code/PDF/DOCX attachments, durable task batches, Search research with source links, model selection, conversation continuation and terminal progress watching. See MCP feature usage.
# Check your installation without starting a browser
npx -y @rnaefe/chatgpt-wrapper@latest doctor
# Reconnect your saved session if login is needed
npx -y @rnaefe/chatgpt-wrapper@latest connect
# Upgrade a managed installation
npx -y @rnaefe/chatgpt-wrapper@latest update
# Remove Wrapper from the selected MCP client's config
npx -y @rnaefe/chatgpt-wrapper@latest uninstallUninstall preserves your local profile and task data so other installed clients are not unexpectedly broken. To remove your data, follow the complete removal instructions.
Question | Answer |
Is this an official OpenAI integration? | No. This is an independent experimental project that automates the ChatGPT web interface, not an official API. |
Does it bypass account limits or verification? | No. Your normal account limits apply; login and browser challenges require you. |
Will it work headlessly? | Headless mode is experimental and has encountered browser verification. Visible mode is the default. |
Is it reliable on Windows/Linux? | Client configuration is supported in documented environments, but equivalent live end-to-end acceptance hasn't been completed. |
Where is my data? | Your profile, task history and local diagnostics stay on your machine. Task prompts and uploaded content are sent to ChatGPT. |
Will it edit my repository? | No. It returns results to the MCP client, which decides what to do with them. |
Security: Treat the saved browser profile and local tokens as credentials. Do not publish them, and only submit data you are authorized to share. Review the provider's applicable terms. Security and privacy · Known browser limitations.
Issues, bug reports and small pull requests are welcome. If a provider UI change breaks a task, please share a minimal reproduction without private prompts, screenshots of your account, browser profiles or credentials.
Open an issue · Contribute · Report a security issue privately
Licensed under GPL-3.0.
This server cannot be deployed
Shared memory for your AI tools: tasks, notes and context every tool can read and write over MCP.
Persistent memory, knowledge graph, task management and more for AI agents via MCP.
Shared memory and mail for your AI agents. Verified with Claude Code; other MCP clients in testing.
Shared memory and actions for Claude, Kiro, OpenAI, Cursor, and other MCP-compatible AI clients.