Assistants write Markdown all day — release notes, specs, summaries — and leave it in a chat. Connect TransformPipe once and the assistant saves each document to your account, where you can open it on any device or send it as a link.
https://transformpipe.com/api/mcp · Full setup guide
Four things that stop being a chore once the assistant can reach your account.
Release notes, a spec, the summary of a long thread: ask for it to be saved and it goes into your account under a title you can search for. Tomorrow it is there on another laptop or on your phone, long after the chat has scrolled away.
“Save this as the Q3 plan.” — Saved · Q3 plan
The assistant saves the document, shares it and answers with the address. Whoever opens it sees a finished page — headings, tables, code — with no account and no asterisks. Share it with anyone who has the link or only with the addresses you name, and revoke it whenever you like.
“Publish it and give me the link.” — Link created · transformpipe.com/s/…
Every saved document is in one list, whichever conversation and whichever assistant it came from. Ask for it by what it was about and the assistant looks it up, instead of you scrolling through forty chats.
“Find the migration spec from last week.” — Found · Migration spec, 16 Sep
When the assistant updates a document, the new text is kept as a version beside the old one. You can see what replaced what, and ask what changed between them.
“Update the spec and keep the old version.” — v1 → v2 · both kept
What happens to a document an assistant wrote, the way most people handle it today and the way it goes with the connector.
| Copy and paste | With TransformPipe | |
|---|---|---|
| Where it ends up | In the chat, or a note you pasted it into | In your account, under a title |
| Finding it later | Scrolling back through conversations | Search, or ask the assistant for it |
| From another assistant | A different app, a different history | The same list, whichever tool wrote it |
| Sending it to someone | Pasted Markdown, asterisks and all | A link to a finished page |
| Who can open it | Whoever you forwarded it to | Anyone with the link, or only the addresses you name |
| After a rewrite | The old text is wherever you left it | Both versions kept, side by side |
One address for every assistant. There is no key and nothing to generate.
In Claude: Settings, Connectors, Add custom connector. Sign in with Google when it asks.
Save, publish, find, update. The document is in your account and opens on any device.
A grant can be limited to reading. That is enforced on the credential itself, so it holds whatever a document the assistant reads tries to tell it.
500 documents and 100 MB an account, 4 MB a document. Reaching a limit refuses the save; nothing is deleted to make room.
One account and one list of documents, whichever of these wrote them. Claude connects today; the rest are being tested, and each gets its own page once connecting it has been seen to work from start to finish.
Custom connector · sign in with Google. claude.ai, Claude Desktop and Claude Code. Add one custom connector, sign in, and ask Claude to save, publish or find a document.
Developer mode connector. ChatGPT takes custom MCP connectors in developer mode, on the plans that offer it. The sign-in is being tried against this server now.
Remote MCP server. Cursor adds remote MCP servers with a sign-in, but returns from it to the editor through its own link scheme, which this server does not accept yet. That is the part being worked on.
Gemini CLI, remote MCP. Gemini CLI adds remote MCP servers with a sign-in through the browser, the same way Claude Code does, and is next to be tried.
Copilot agent mode, remote MCP. GitHub Copilot’s agent mode in VS Code connects to remote MCP servers and signs in through the browser. Being tested.
Remote MCP server. Windsurf’s assistant takes remote MCP servers too. Whether its sign-in completes against this server is still to be checked.
MCP, the Model Context Protocol, is how an AI assistant calls tools outside the chat. TransformPipe runs an MCP server at `/api/mcp`; adding it to your assistant as a custom connector gives it a handful of tools — save a document, share it as a link, list and open what you saved, keep versions — that it uses when you ask in plain words.
Yes, with no plan to pick and no card to enter. An account holds 500 documents and 100 MB of Markdown, and a single document can be up to 4 MB. Reaching a limit refuses the save and says so rather than quietly deleting an older document to make room.
No. Adding the connector in your assistant sends you through a normal sign-in, and that sign-in is the account — there is no separate registration, no API key to generate and nothing to paste. The same account works on the website, so the documents your assistant saves are there when you sign in on another device.
Ask for it in the same conversation: “save this”, or “save this as the Q3 plan”. The assistant sends the Markdown it wrote to your account, where it gets a title, a place in a searchable list and a version history — instead of staying in a chat you would have to scroll back through or copy out by hand.
Ask the assistant to publish it. It saves the document, shares it and answers with the address: a finished page with headings, tables and code rendered, not raw Markdown with its asterisks. Share it with anyone who has the link, or make it a read-only document link that only the email addresses you name can open, and revoke it at any time.
No. Documents you save are not read by us and they are not used to train any model. They are stored in your account, reachable by you and by the assistants you connected, until you delete them — and deleting one removes it rather than hiding it.
Only one document at a time, and only with an explicit confirmation in the same request, so a vague instruction cannot empty an account. A read-only grant can list, open and summarise your documents but cannot save, share or delete anything, and that limit is enforced on the credential itself rather than on what the assistant is told.
The assistant loses access immediately: removing the connector revokes the grant. Your documents stay in your account, and the links you already shared keep working until you revoke them yourself. Connecting again later picks up the same account and the same list of documents.
No. A shared document is read-only for whoever opens it: they can read the page and download it, but they cannot change it, delete it or pass the share on. Documents shared with your address by other people appear in your own list, marked as shared with you.
No. TransformPipe is an independent MCP server, not a product of Anthropic, OpenAI, Google or any editor vendor. Each assistant connects to it the way it connects to any custom connector, using the standard sign-in the protocol defines.
Your next document is being written in a chat right now. Put it somewhere it will still be next week.