wordcraft/docs/control-protocol.md

Control protocol

Start the app with a control port: wordcraft --control 7981 (or WORDCRAFT_CONTROL_PORT=7981). A missing or invalid port after --control (not 1–65535) prints an error and exits with status 2. It listens on 127.0.0.1 only. Send one JSON object per line, with the window's key (see Keys); each gets one reply line:

{"id": 1, "key": "…", "method": "engine.execute", "params": {"command": "text.insert", "params": {"text": "Hello"}}}
{"id": 1, "ok": true, "result": {"anchor": {...}, "focus": {...}}}

Errors come back as {"id": …, "ok": false, "error": "message"}. A failed command leaves the document unchanged. A line that isn't valid JSON gets one error reply, then the server closes the connection (so an HTTP request sent to the port can't run a command); reconnect to continue.

Keys

Any local program (and possibly a web page) can reach a loopback port, so every request needs the window's key. Once the port is bound, the app writes a random key (256 bits, as hex) to <settings>/control-key.<instance> and removes the file when the window closes; a new start writes a new key. WORDCRAFT_PROFILE_DIR overrides the whole profile folder for isolated runs, including recovery copies. See local recovery. <settings> is the folder that holds the preferences: on Linux $XDG_CONFIG_HOME/wordcraft (by default ~/.config/wordcraft), on macOS ~/Library/Application Support/WordCraft, on Windows %APPDATA%\WordCraft. <instance> is the Flatpak instance id inside Flatpak, and local-<port> otherwise. On Unix the file has mode 0600.

cat ~/.config/wordcraft/control-key.local-7981   # Linux, outside Flatpak, port 7981

• Send the key as top-level "key" with every request, or once per connection with {"method": "auth", "key": "…"} (reply {"ok": true, "result": {}}); later requests on that connection may leave key out. Only auth reads params.key; for other methods params.key is an ordinary parameter (ui.key {"key": "Enter"} presses Enter).
• The key is checked on every request. A missing or wrong key gets {"ok": false, "error": "unauthorized"}, and the app closes the connection.
• A line over 1 MiB gets {"ok": false, "error": "line too long"}, and the app closes the connection.
• A connection that sends no authorised request within 30 s of being accepted is closed. That is one limit: blank lines and bad JSON do not extend it. After an authorised request the connection may stay idle.

wordcraft-cli mcp --connect takes the key from WORDCRAFT_CONTROL_KEY, or else from the key file for the port it connects to, and reads it again on every new connection (the app may have restarted). It sends the key only to loopback addresses (127.0.0.0/8, ::1, localhost). For any other host it sends no key, says so, and the app refuses its requests: to drive an app on another machine, forward its port to 127.0.0.1 (for example ssh -L 7981:127.0.0.1:7981 host) and set WORDCRAFT_CONTROL_KEY to that app's key.

Methods

Method	Params	Result
engine.execute	command, params	the command's result (any id from engine.commands, including UI commands ui.tab, ui.dialog ({"name": dialog, "part"?: table style part}; part opens New / Modify Table Style (newTableStyle / modifyTableStyle) on that part: wholeTable, headerRow, firstColumn, bandedRows, nwCell… as table.modifyStyle names them), ui.backstage, ui.zotero.* (see docs/zotero.md), ui.language (`{"value": "auto"
engine.execute	command, params	the command's result (any id from engine.commands, including UI commands ui.tab, ui.dialog ({"name": dialog, "part"?: table style part}; part opens New / Modify Table Style (newTableStyle / modifyTableStyle) on that part: wholeTable, headerRow, firstColumn, bandedRows, nwCell… as table.modifyStyle names them), ui.backstage, ui.zotero.* (see docs/zotero.md), ui.language (`{"value": "auto"
engine.commands	—	every command: id, label, location, shortcut, params, enabled
document.inspect	text?	blocks (text, style, runs, lists, tables), parts, sections, selection, pages
ui.inspect	—	UI state, view, dialog, window size, page rects on screen, caret, comment balloons (balloons: id + screen rect; balloon: the selected one), perf
ui.click	x, y, button?, count?, shift?, cmd?, alt?	real pointer input at window coordinates
ui.clickText	page (0-based), x, y (points from the page's top-left), count?	click inside a page
ui.move / ui.drag	x,y / x,y,toX,toY,steps?	pointer move / drag
ui.press / ui.release	x,y, modifiers	primary button down / up there (with ui.move between: a drag you can screenshot halfway)
ui.wheel	x, y, dy, dx?, unit? (point/line/page), phase? (start/move/end), modifiers	mouse-wheel or trackpad scroll at window coordinates (dy < 0 scrolls down)
ui.key	key (B, Enter, Left…), shift?, alt?, cmd?	key press through egui (shortcuts apply)
ui.text	text	typed text through egui
ui.screenshot	path?	PNG of the window: written to path, or without one returned as base64 png in the reply
ui.render	path?, page? (1-based), scale?	PNG of a page, rendered by the engine: written to path, or without one returned as base64 png in the reply (never a fixed name in the shared temporary folder)
ui.parity	—	feature-catalog parity
ui.resize / ui.focus / app.quit		window control

Any other method name is treated as a command id: {"method": "format.bold"} works.

The CharProps and ParaProps that format.set, para.set and styles.modify take are listed in MCP server.

Positions are {"story": "body" | {"part": n}, "path": [block, row, cell, block, …], "off": byteOffset}.

Closing the document

engine.execute with file.close never shows the Save / Don't Save / Cancel prompt. On a document with unsaved changes it replies with an error (the app keeps running and the edits stay); save first (file.save), or pass {"force": true} to discard the changes and close. The bare form {"method": "file.close", "force": true} (a method name that is a command id) is guarded the same way, and only a boolean true counts as force. A document without unsaved changes closes at once (#1268). Closing closes the document, not the app: the window stays and shows no page, and file.new or file.open goes on from there (#1712).

Native printing

engine.execute can call ui.printSettings with {} to inspect the printer list, selected printer, copies, duplex, grayscale, discovery/submission status and backend receipt/error. {"refresh":true} starts background discovery; poll the same command until discovering is false. Set printer to a discovered name, integer copies from 1 to 999, duplex to printerDefault, off, longEdge or shortEdge, and boolean grayscale. Invalid updates are rejected without changing earlier settings. ui.printSubmit explicitly submits the current print-ready PDF without opening a dialog; inspect ui.printSettings until busy becomes false and check status/detail. The browser keeps its own print hook. PDF export remains separate.

ui.printSubmit / ui.print also accept the upstream PDF page settings (range, pages, parity, copies, collate). Native printing renders the selected pages once and passes copies/collation to the spooler; the browser and file.printPdf keep copies in the PDF. Omitting native copies uses the configured ui.printSettings value. collate defaults to File › Print's setting; invalid ranges submit no job.

engine.commands {"search":"print"} lists the engine print commands and the app-owned ui.printSettings, ui.printSubmit, and legacy ui.print alias, with current enablement and their parameter documentation. Search is case-insensitive across IDs, labels, locations and parameter descriptions. Listing commands never discovers printers, submits a job or opens a dialog. Settings inspection is always available; submission becomes enabled when the native printer is ready, or when the browser host provides its print flow.