MCP needs wordcraft-cli on your computer. The browser editor on this site can’t be driven by an agent; for AI help in the browser, use the AI writing assistant.
Get the CLI
wordcraft-cli 0.5.0 comes with the Windows installer and portable zip and with the Linux packages, and as a separate signed and notarized download for macOS (wordcraft-cli-0.5.0-macos-universal.zip). The MCP server is built in.
Set up with Claude Code
For headless documents (no window):
claude mcp add wordcraft -- wordcraft-cli mcpTo drive the running desktop app, including screenshots, clicks and keys:
wordcraft --control 7981 &
claude mcp add wordcraft-app -- wordcraft-cli mcp --connect 127.0.0.1:7981Or add it to your agent’s MCP configuration (.mcp.json in a project for Claude Code):
{
"mcpServers": {
"wordcraft": {
"command": "wordcraft-cli",
"args": ["mcp", "--author", "Claude (copyedit)"]
}
}
}What the agent gets
| Tool | What it does |
|---|---|
list_commands | Every command, filtered by a query |
execute, batch | Run one command, such as insert.table with rows and columns, or several in order |
new_document | Blank, sample, letter, résumé or report |
open_document, save_document | docx, doc (open only), odt, rtf, html, md, tex, txt and JSON; saving also writes PDF and PNG |
type_text, select_text | Type at the caret; select the n-th occurrence of some text |
get_text, inspect_document | Read the document back to check the work without screenshots |
render_page | A page as a PNG the model can look at |
screenshot, click, key, ui_inspect | Desktop app only, through --connect |
The server also offers the open document and the command list as resources. Macros can record a sequence of commands and play it back.
Prompts to try
- “Open
proposal.docx, turn on Track Changes, tighten the executive summary and save it asproposal-edited.docx.” - “Make a one-page letter from the letter template to the address in
client.txt, then save it as PDF.” - “Convert every .docx in
reports/to PDF and list any document whose first heading isn’t Heading 1.”
Without MCP: the CLI on its own
wordcraft-cli convert report.docx report.pdf # docx, pdf, odt, rtf, html, md, tex, txt, png
wordcraft-cli text report.docx # plain text
wordcraft-cli info report.docx # pages, words, paragraphs, properties (JSON)
wordcraft-cli inspect report.docx # structure as JSON
wordcraft-cli run --template sample \
--cmd 'select.text={"text":"Membership"}' --cmd format.bold --save out.docx
wordcraft-cli commands # every commandFAQ
Can an AI agent control the browser version on this site?
No. The web build has no control channel; the MCP server is part of wordcraft-cli and runs on your computer. For AI help with a document in the browser, use the AI writing assistant panel; for agent automation, use the CLI or the desktop app.
Which agents work with it?
Any MCP client that can launch a local stdio server: Claude Code, Claude Desktop, Cursor, VS Code agents and others. They all take the same command and arguments.
What is the difference between headless and --connect?
Plain wordcraft-cli mcp works on documents without a window: fast, and fine for writing, converting and checking files. With --connect it drives a running WordCraft desktop app instead, so the agent can also take screenshots, click and press keys, and you watch the document change.
Is the control port open to the network?
No. The desktop app listens only when you start it with --control, only on 127.0.0.1, and every request needs the random key the window writes when it starts. The MCP bridge sends that key only to loopback addresses.
How do reviewers tell the agent’s edits from mine?
Start the server with --author, for example --author "Claude (copyedit)", and turn on Track Changes with review.trackChanges {"value": true}. Every insertion, deletion and comment the agent makes is then credited to that name.
Source: the WordCraft README, MCP server notes and control protocol. Flags and tools can change between releases; check wordcraft-cli --help and wordcraft-cli commands.