Three sources, one landing place
Claude output is not one format, so it helps to be precise about which one you have.
- A chat answer — Markdown under the renderer. Paste it; headings, lists, tables and math come across as structure.
- An artifact — a self-contained HTML preview. Copy the text, or export the artifact as a file and drop it in; the parser reads the blocks rather than the page’s CSS.
- A file Claude Code wrote — a .md on disk. Open it in JotDoc, or paste it; either way it becomes a document you can edit by block.
Keeping Claude in the loop
The usual reason people stop using a chat draft is that the document and the conversation drift apart. The chat remembers a version; you fixed three paragraphs in Word; nobody knows which is current.
Connect JotDoc as an MCP server and Claude reads the latest text of the document, not a copy from earlier in the thread. It proposes a diff and you accept or reject each change in the document. If you only need one paragraph improved, Hand off copies a single sentence for you to paste — no connector required.
What you get at the end
Export → Word writes a .docx with heading styles, real tables, footnotes and equations Claude can write but cannot itself export. From there a reviewer opens a link, comments on the sentence they mean, and you revise in place — no account, no round-tripping files.
Questions this page gets
Does Claude need a special plan to connect?
No. MCP is a protocol, not a tier. Claude, Claude Desktop, Claude Code and any MCP-capable client can connect to the JotDoc endpoint; you authorize once in the browser and can revoke it later.
Can Claude edit the document directly?
It can propose changes, not apply them. Every edit lands as a suggestion you accept or reject. That is enforced by the server, not by asking the model politely.
What happens to my token usage?
It is your account and your tokens throughout — a free Claude account is enough. JotDoc does not proxy the model and does not spend credits on the AI collaboration path.
Where to go next
- AI apps over MCP — how the AI reads the latest text and why it can only suggest changes.
- Peer review without accounts — the step between “finished” and “sent”.
- Connect an AI app — five minutes, one endpoint.
- Working on a Mac? The desktop app opens the same documents from Finder.