Offline Protocol plugin for ChatGPT and Codex
With the Offline Protocol plugin installed, you can ask ChatGPT or Codex to plan an app built on Offline Protocol, choose compatible packages and walk through the integration guide for your workflow. The plugin is listed in the plugin directory at chatgpt.com/plugins, and the Start building section of the Offline Protocol homepage links to it.
OpenAI’s documentation describes the install steps:
- Open the plugin’s details in the directory.
- Select the plus button to install it.
- If the plugin asks to connect its MCP server, connect it when prompted.
- Start a new chat and ask ChatGPT or Codex to use the plugin.
OpenAI describes a plugin as a package that can bundle an MCP connection with skills, which are reusable instructions. ChatGPT and Codex share one plugin catalog, so the same plugin is discoverable from both. Plugins work in Chat and Work in ChatGPT on the web, desktop and mobile, and in Codex in the ChatGPT desktop app. In Codex CLI, /plugins opens the plugin browser; start a new session after installing.
In ChatGPT on the web, plugin tools run on a remote MCP server, according to OpenAI. That suits the planning side of the work. When you want Codex to write files into your repository, add the local server described next; the docs keep the file-writing tools in local mode.
Codex local MCP server setup
Codex takes the local MCP server with one command, run once:
codex mcp add offline-protocol -- npx -y @offline-protocol/cli@0.2.9 mcp serve
Or add this to ~/.codex/config.toml:
[mcp_servers.offline-protocol]
command = "npx"
args = ["-y", "@offline-protocol/cli@0.2.9", "mcp", "serve"]
The server runs on your machine and talks to Codex over standard input and output. It needs Node.js 18 or later and no login or API key. The first launch downloads that CLI release from npm, and later launches use the npm cache.
OpenAI’s documentation adds three details that help here. The ChatGPT desktop app, Codex CLI and the IDE extension share this configuration, so you set it up once. codex mcp list shows the configured servers, and /mcp in the Codex terminal UI shows the active ones. A project-scoped .codex/config.toml also works, in trusted projects only.
Hosted Offline Protocol MCP in Codex
Codex can also use the hosted endpoint, which runs without a local install and offers the 15 read-only tools; scaffolding runs in local mode. It takes an app API key, created in the developer portal, and the matching App ID. Organization keys are rejected. In ~/.codex/config.toml, with OFFLINE_API_KEY set in the environment used to start Codex:
[mcp_servers.offline-protocol]
url = "https://mcp.offlineprotocol.com/mcp"
bearer_token_env_var = "OFFLINE_API_KEY"
http_headers = { "x-app-id" = "app_your_application_id" }
Configure one mode under the offline-protocol name, not both.
Offline Protocol tools in ChatGPT and Codex
Connected to the local server, Codex can plan the architecture of an app, pick compatible packages, read the integration guide for a workflow and scaffold a starter project. The local server provides 18 tools for packages, capabilities, workflows, examples, skills, templates, planning and scaffolding.
| ChatGPT plugin | Codex, local MCP | Codex, hosted MCP | |
|---|---|---|---|
| Install | Plugin directory | codex mcp add or config.toml | config.toml |
| Where the tools run | A remote MCP server | Your machine | Offline Protocol hosted service |
| Plan, choose packages, read guides | Yes | Yes | Yes |
| Scaffold files in your repository | Use Codex with local MCP | Yes | Use local MCP |
The planning and lookup tools include generate_architecture, resolve_packages, get_skill, get_workflow and preview_scaffold. The three scaffolding tools, scaffold_project, integrate_packages and init_project, write files and run package-manager commands with your permissions. Codex lets you set a server’s default_tools_approval_mode, and OpenAI’s documentation says its writes value prompts for tools that aren’t marked read-only.
The ready workflows local-handoff, backend-delivery and nearby-service each scaffold a React Native app. If you are still choosing a package for another platform, Which Offline Protocol SDK should I use? compares them.
A first Offline Protocol request in ChatGPT or Codex
A concrete request gets a concrete plan. The Offline Protocol docs suggest a first request like this:
Add an encrypted local handoff to this React Native app. Explain the device requirements, choose compatible packages, and show how the receiving app will persist the handoff.
Add your repository, target device models and OS, installed SDK version and intended workflow. The live agent setup prompt is the canonical setup instruction; give Codex its URL rather than a copied version.
Next steps after Codex scaffolds the app
Once Codex has planned and scaffolded the integration, the next steps happen on your machine and your devices:
- Test on the target devices. The docs ask you to compile the changed application and test on the devices, where radio delivery is confirmed.
- Review the generated changes with the rest of your code before you commit them.
- Keep keys in the environment. Local mode needs none, and the hosted configuration reads the key from
OFFLINE_API_KEY.
The same server works in Claude, Cursor and VS Code. For how its local, hosted and public modes differ, read What is the Offline Protocol MCP server?