Connect ChatGPT, Codex, Claude, Gemini and Grok

You decide which devices the assistant may use. Ask it to send a file to one of your devices, to a person, a server or your pipeit Cloud, to check a transfer or to tidy up your Cloud storage. Every action is shown to you first and confirmed by you.

MCP server address, copy it completely:

https://mcp.pipeit.now/mcp

Connection: Streamable HTTP. Sign-in: OAuth, through your pipeit account. Leave client ID and client secret empty: the client registers itself. Do not copy API keys or sign-in codes into a chat or a configuration file.

Claude and Claude Desktop

1. In Claude open Customize → Connectors.
2. Use + → Add custom connector to create your own connector.
3. Enter the name pipeit and the server address above.
4. Leave the advanced OAuth fields empty. Connect and finish the sign-in with pipeit. Check the return address and release the devices you want.
5. In the chat, switch pipeit on through + → Connectors.

On team and enterprise plans an organization owner may have to create the connector first. It is a remote connector; no local desktop extension has to be installed. Source: Anthropic: custom remote connectors (Opens an external website). Menu names can vary with the client version.

ChatGPT

1. If your account offers it: Settings → Security and sign-in → Developer mode, switch it on.
2. In the plugin or app management, add your own MCP connection with +.
3. Enter the name pipeit and the complete server address. Use OAuth as the authentication; no fixed client ID and no secret.
4. Sign in to pipeit and release devices deliberately. Then choose the connection in the tools menu of the chat you want.

Access to your own connections depends on your account and organization policies. Source: OpenAI: connect MCP in ChatGPT (Opens an external website).

Claude Code

Set it up once for your user, so it works across projects:

claude mcp add --transport http --scope user pipeit https://mcp.pipeit.now/mcp

Then open Claude Code, call /mcp, choose pipeit and finish the sign-in that is offered. If the connection should only exist in one project, use --scope project instead. Each person signs in with their own pipeit account. Source: Claude Code: MCP and OAuth (Opens an external website).

Codex

codex mcp add pipeit --url https://mcp.pipeit.now/mcp
codex mcp login pipeit --scopes targets:read,transfers:read,offline_access

Finish the browser sign-in and the release. If a sign-in was already offered and completed when you added it, no second login is needed. Then open a new Codex session. Check an existing MCP configuration named pipeit before changing it; the guide does not install a second local file client. Source: OpenAI: MCP in Codex (Opens an external website).

Gemini and Grok

1. In the MCP or connector settings of Gemini or Grok, add a custom MCP server with the name pipeit and the server address above.
2. Choose OAuth as the sign-in; leave client ID and client secret empty.
3. Sign in with your pipeit account and release the devices you want. Then switch pipeit on in the chat you want.

What you can try

• “Which devices have I released for pipeit?”
• “Send this file to my laptop.”
• “Open pipeit so I can prepare a transfer.”
• “Check the status of this pipeit transfer: [transfer ID].”

With pipeit Cloud the assistant can also list your Cloud spaces, folders and versions and search by name (list_cloud_spaces, list_cloud_folder, get_cloud_node, list_cloud_versions, search_cloud). With a separate permission it can search content and show previews (search_cloud_content, get_cloud_preview). Every call is limited to the Cloud spaces you select; permission for metadata alone gives no contents. Cloud files are not end-to-end encrypted, and snippets or previews can reach your AI provider. Sending (prepare_transfer, send_file, cancel_transfer) always shows the planned transfer first and asks for your confirmation; files on a device are sent through the pipeit app on that device (local_agent_required), and URLs are not accepted as a source (unsupported_file_source). The connector never reads your local file system, chat files, keys or shell.

An empty device list does not mean the sign-in failed: you can connect first without device access. Under Connected applications in pipeit you can release single devices or revoke rights later. A release applies for the duration you choose, at most 30 days. Switching it on for a chat does not widen any device permission.

Help

• The URL shows a method error in the browser: /mcp is the interface for the AI client, not a web page. Enter the whole URL in its MCP configuration.
• Sign-in is missing: choose OAuth and connect again. Do not copy bearer tokens into the chat by hand.
• No devices visible: release devices explicitly under “Connected applications” in pipeit.
• Transfer not visible: check the concrete transfer ID, the status permission and the device release. There is no general read access to the account.
• Too many requests: wait for the time shown; do not register again and again.
• I cannot add a custom connector: check availability and organization policies with the respective platform.
• Remove the connection: revoke the application in pipeit, then remove it in the AI application.