An assistant writes Markdown all day and cannot hand you a page. Connecting this one lets it do both, without anybody copying text between tabs.
TransformPipe runs an MCP server at `/api/mcp`. MCP is the protocol assistants use to call tools, so adding the address as a connector gives the assistant a set of verbs it can use on your behalf: convert this, save it, share it, list what is there.
There is no key to paste. Adding the connector sends you through a normal sign-in, and the assistant is granted access to that account until you disconnect it — the same shape as signing into any other application with your account.
On claude.ai, TransformPipe is listed in the connector directory at `claude.ai/directory/tp`: open it, press Connect, and sign in when asked — the tools appear in the next conversation. Where the directory is switched off, Settings, then Connectors, then Add custom connector takes the same server by address, `https://transformpipe.com/api/mcp`.
From a terminal, one command does the same thing: `claude mcp add --transport http transformpipe https://transformpipe.com/api/mcp`.
In ChatGPT, TransformPipe is added by address as an MCP app: open Plugins, press Add, choose Create MCP App, give it `https://transformpipe.com/api/mcp`, and sign in when asked. Documents then come back as cards in the chat.
Nothing else is configured. The connector can be removed from the same screen, and removing it revokes the access immediately.
Nineteen tools, all named `tp_`. The ones that matter in a conversation are `tp_convert_from_markdown`, which turns Markdown into a finished HTML document, `tp_convert_to_markdown` for a file going the other way, `tp_save_document`, which keeps the result in your account, and `tp_share_document`, which publishes it and returns a link you can send — or adds or removes one reader.
The rest are the ones an assistant reaches for on its own: `tp_list_documents` and `tp_get_document` to find something you made earlier, `tp_summarize_document` to say what a long one contains, `tp_update_document` to change one and keep its link, `tp_document_versions` to show what replaced what, `tp_list_folders`, `tp_create_folder`, `tp_update_folder`, `tp_delete_folder` and `tp_move_documents` to keep documents in folders, `tp_share_folder` to share a whole folder the way a document is shared, `tp_usage` to check how much room is left, `tp_delete_document`, and `tp_search_documentation`, which answers questions about TransformPipe from its documentation.
In practice the useful sentence is short. Ask it to write the release notes, then ask it to publish them — the assistant converts, saves and shares, and answers with the address.
The connector acts as you, in your account, on documents you own. It cannot change the account, read your password, create API keys, or reach anybody else’s documents.
A grant can also be read-only, in which case the assistant can list, fetch and summarise but cannot save, share or delete — and that restriction is enforced on the credential itself rather than on the tools, so it holds whatever the assistant asks for.
Worth knowing rather than worrying about: an assistant with a connector is standing authority to act, and a document it reads can contain instructions aimed at it. That is the honest cost of the convenience, and the reason a read-only grant is the right default for anything you have not written yourself.
A connector suits the document that exists inside a conversation and nowhere else. For a file already on disk, dropping it on the converter is quicker; for something that happens on every merge, the API or the GitHub Action is the right shape; and for a folder of four hundred files, a local converter beats a conversation.