attach_add
ChatGPTSave a file as a document attachment, or replace an existing attachment. Attachments are most often images, for example a photo used on a slide, but may also be PDFs, audio, video, spreadsheets, or other files. Returns the actual saved filename in structuredContent.name and response text. doc-id is the USolver document ID returned by doc_create or doc_find. It identifies the document whose attachments will be changed. Provide exactly one of file, url, data, or svg. A file is a host-provided reference with download_url and file_id, plus optional file_name and mime_type. Both file and url are downloaded into document storage. data accepts base64 or a data URL; svg accepts SVG markup. The maximum file size is 10 MiB. For raw base64, supply a filename with an extension or mime_type to identify the format. Omit replace or set it to false to add a new attachment. The optional name is a preferred filename. If it conflicts with an existing attachment, a unique filename is returned. To replace an attachment, set replace: true, supply its exact returned name including extension, and provide the new content. The name must exist, and the replacement must use the same file format. This updates the attachment without creating another entry or slide element. All existing uses retain their placement and refer to the updated file. This tool manages stored attachments only. For example, after adding an image and receiving photo.png, use elem_add with an image element whose background.image is photo.png to place it on a slide. Use attach_remove with that name to delete the attachment.
attach_remove
ChatGPTRemove an attachment from a document by its exact name, including extension. Attachments are most often images, but may also be PDFs, audio, video, or other files. doc-id is the USolver document ID returned by doc_create or doc_find. It identifies the document whose attachments will be changed. Deletes the stored file and its attachment entry, and synchronizes the removal to cloud storage. Returns the removed name. A name that is not in the document's attachments returns an error. This does not delete or edit slide elements. Any existing references to the removed file will no longer resolve. To replace a file while keeping its uses, use attach_add with name and replace: true instead.
doc_create
ChatGPTCreate a USolver presentation with its slides, backgrounds, and elements. Image elements can be included in the slides or added later with elem_add, using the attachment name returned by attach_add as background.image. For a single title slide, the text parameter supplies the title. The result includes the document ID for later editing and an edit link. Theme selection Every document gets a theme. A theme is a design system: slide designs (background and decoration for title, section, content and closing slides), a palette, fonts, a type scale and styles for element roles (titles, cards, callouts, tables, charts, buttons). - If the user named a theme or described a look, use it: a built-in id or a saved custom theme id (my-...; if you do not know it, look it up with list_themes, which shows nothing to the user). theme takes ids only: a new custom theme is made first with the custom-theme skill (ask what the user wants, preview it with theme_preview, save it with theme_save once they confirm), then pass the my-... id. The document keeps a copy of its theme. - Otherwise do not ask which theme they want: pick the built-in theme that fits the content, audience and tone best and create the document right away. The preview has theme buttons (previous, a Theme menu, next), so the user can switch the theme in one click. In your reply name the theme you picked and mention that it can be changed in the preview. - Show the theme picker (theme_picker) only when the user asks to see or choose themes, and build a custom theme only when they ask for one or describe a look none of the built-in themes gives. Built-in themes: - aurora Aurora (dark; Creative, Tech): Vivid violet-to-magenta gradients behind frosted-glass cards, with modern Sora type. Launches, keynotes and inspiring talks. - blueprint Blueprint (dark; Tech, Education): Engineering-blue grid paper, monospace headings, white linework and amber highlights. STEM lessons, engineering and technical talks. - boardroom Boardroom (light; Business): Confident navy and white with a teal highlight, Manrope headings and crisp tables. Business reviews, reports and proposals. - botanical Botanical (light; Creative, Editorial): Sage greens on warm cream with an elegant Playfair voice and soft organic shapes. Nature, wellbeing, biology and design. - chalkboard Chalkboard (dark; Education, Kids): A green chalkboard in a wooden frame, handwritten Caveat headings and chalky pastel accents. Made for classrooms and lessons. - graffiti Graffiti (dark; Creative): Spray-painted lettering on a dark concrete wall, neon pink and acid-lime splatter with paint drips. Loud and rebellious: youth events, music, street culture, creative pitches. - midnight Midnight (dark; Tech): Deep navy with an indigo glow, techy Space Grotesk headings and electric cyan accents. Software, data, science and product talks. - paper Paper (light; Education, Editorial): Warm paper, a soft Fraunces serif and terracotta accents. Calm and literary: humanities, essays, storytelling and lectures. - playground Playground (light; Kids, Education): Sunny, rounded and colorful: bubbly Baloo type, soft cards and confetti shapes. For young learners and playful topics. - pop Pop (light; Creative): High-energy color blocking, chunky Bricolage headings and hard drop shadows. Bold, punchy and impossible to ignore: pitches, marketing, workshops. - swiss Swiss (light; Minimal, Business): Crisp white, bold Inter type, one signal-red accent and a strict grid. Neutral and confident: works for almost any topic. Build slides so the theme can do its job: - Give each slide a kind: title for the first slide, section for slides that open a new part, content for regular slides, closing for the last one. Do not set slide backgrounds unless asked. - Give elements a theme-role (title, heading, subtitle, body, caption, quote, stat, card, callout, accent, badge, divider, button, image, table, chart) and leave out colors, fonts and font-size so the theme provides th…
doc_find
ChatGPTFind USolver documents by full or partial title. The search is limited to documents owned by the authenticated user. Returns Matching document titles, document IDs (doc-id), and timestamps. A returned doc-id can be passed to doc_read to load its slides and elements.
doc_preview
ChatGPTDisplay an interactive preview of the latest saved version of a USolver document. An optional slide selects which slide is shown first. Behavior - The preview reflects the document state at the time it is requested. - A document with a declared image source must contain a matching image before it can be previewed. - It is typically called once, after all changes for a request have been made. - This tool is read-only and does not modify the document. Theme buttons For a themed document the preview has theme buttons next to "Edit in USolver": previous theme, a Theme menu with all themes, and next theme. A theme chosen there is shown in the preview at once and saved to the document a second after the user stops choosing; you are told about the change. Create new theme... in the menu asks you in the chat for a custom theme: follow the custom-theme skill (ask what they want, show it with theme_preview, save it with theme_save once they confirm), then apply the returned id with doc_theme and preview the document again.
doc_read
ChatGPTRead an existing USolver document by doc-id for inspection or editing. Returns - Document title and canvas size - Every slide and its visual elements - Complete saved element objects, including all properties and nested data Use doc_preview to display the saved document interactively.
doc_theme
ChatGPTRestyle an existing document with another theme, given by id: a built-in theme or a saved custom theme (my-...). A custom theme must be previewed with theme_preview, confirmed by the user and saved with theme_save first. Only documents created with a theme (every document made by doc_create) can switch themes. The document keeps a copy of the theme, so later changes to or removal of a saved theme do not affect it. What changes - Slide backgrounds and theme decoration shapes, per slide kind (title, section, content, closing). - Colors, fonts, borders, corner radius and shadows of elements that have a theme role (titles, text, cards, callouts, tables, charts, buttons, images). - Values that were set explicitly or changed by hand (anything that differs from what the old theme set) are kept. - Sizes, positions, text and data never change. Font sizes stay the same, so check slides with long text after switching to a theme with a wider font. Show the result with doc_preview.
elem_add
ChatGPTAdd one or more elements to an existing slide. Supported elements include text, buttons, inputs, images, shapes, lines, and charts. Slide Pass slide with the id of the slide to add the elements to (such as "slide2", as returned by doc_read or slide_add) or its 0-based index. If it is omitted the elements are added to the first slide. All elements in one call go to the same slide. Returns The IDs and names of the new elements. These identifiers can be passed to elem_update. Element types by group General label, input, button, toggle, image, slider, timeline2, timer, radio, check, list, drop, video, audio, minislide, quiz, webview, table, flashcard, hotspot, canvas2d Lines & Geometry shape-path, line, point, circle, ellipse, arc, triangle, rectangle, polygon, star Groups group Gages gage-linear, gage-circular, gage-radial Charts & Plots fx-plot, chart-general, chart-bar, chart-scatterplot, chart-pie, chart-radar, chart-polar, chart-histogram, chart-contour, chart-bullet, chart-box, chart-surface, chart-pictograph, chart-wordcloud, chart-parallel, chart-heatmap, chart-layouttree, chart-timeline, chart-chord, chart-sankey Element properties Coordinates use slide pixels. The default canvas is 1280 × 720, with the origin at the top-left. Properties may use dot-path keys or nested objects. Common left, top, width, height, name, show-when, and show-hidden-outline (true: while the element is hidden by show-when, a dashed outline filled with its background color at 20% shows where it will appear. In a progressive reveal set it on one element per step only: the box the step's other elements sit on, see the slide properties). Positioning Elements are not auto-laid-out: left/top/width/height must be computed deliberately against the canvas size, not guessed. Align elements to the slide edges and to each other — matching left edges for a stacked column, centering with left = (canvas width − element width) / 2, equal gaps between siblings — so nothing ends up crowded, overlapping, or visibly off-grid. Labels default to text-align: "center". That suits short titles but is usually wrong for body text, bullet lists, or captions — set text-align (e.g. "left") explicitly whenever centered text would look out of place. Text For labels, buttons, inputs, checks, radio controls, drop-downs, and lists: - text; labels accept HTML and <br> for line breaks - color, font-family, font-size (pixels), font-weight, and font-style - text-align, vertical-align, line-height, and padding.all Size the box to comfortably fit the text, not just match it: estimate the rendered text's width and height from font-size, character count, and line-height (accounting for wrapping), then make width/height larger than that estimate. A box sized exactly to the text leaves no room for padding and risks clipping or crowding it, especially with longer or multi-line text. Set padding.all (commonly 12–24 px, more for larger text) so the text never touches the box edges. Backgrounds For slides, labels, buttons, inputs, images, webviews, charts, and similar elements: - background.type: solid, gradient, or an empty string for none - background.color, background.image, and background.size - background.gradient.type: linear, circle, or ellipse - background.gradient.stops: an array of {color, stop} values - Linear gradient coordinates x1, y1, x2, and y2 use percentages Borders and shadows - Border: border.style.all, border.width.all, border.color.all, and border.radius.all - Shadow: box-shadow.display, box-shadow.color, box-shadow.offset-x, box-shadow.offset-y, and box-shadow.blur Buttons Use text and button-style. Supported styles are gray, blue, green, orange, red, and custom. Custom buttons use background.color. Shapes and lines Shapes include rectangles, circles, ellipses, polygons, stars, paths, and lines. - Fill and stroke: fill.color, fill.type, stroke.color, and…
elem_delete
ChatGPTDelete one or more elements from an existing document. Elements may be identified by ID or name.
elem_update
ChatGPTChange properties of one or more existing elements, including text, position, size, colors, fonts, borders, and backgrounds. Element references Pass slide (a slide id such as "slide2" or a 0-based index). It is required, and every element in updates and order must be on that slide. Reference each element by ID, such as "label-3", or by name, such as "TXT1". These identifiers are returned by doc_read and elem_add. Names are looked up on the given slide only. Reordering Pass order to change the stacking order of the elements of slide: a list of element IDs or names, first = bottom, last = top. The slide's element list is rebuilt in exactly this order, after any updates in the same call. The list must contain every element of that slide exactly once. Members of a group must stay together, directly after their group element (see group-general). updates is optional when order is given. Element types by group General label, input, button, toggle, image, slider, timeline2, timer, radio, check, list, drop, video, audio, minislide, quiz, webview, table, flashcard, hotspot, canvas2d Lines & Geometry shape-path, line, point, circle, ellipse, arc, triangle, rectangle, polygon, star Groups group Gages gage-linear, gage-circular, gage-radial Charts & Plots fx-plot, chart-general, chart-bar, chart-scatterplot, chart-pie, chart-radar, chart-polar, chart-histogram, chart-contour, chart-bullet, chart-box, chart-surface, chart-pictograph, chart-wordcloud, chart-parallel, chart-heatmap, chart-layouttree, chart-timeline, chart-chord, chart-sankey Element properties Coordinates use slide pixels. The default canvas is 1280 × 720, with the origin at the top-left. Properties may use dot-path keys or nested objects. Common left, top, width, height, name, show-when, and show-hidden-outline (true: while the element is hidden by show-when, a dashed outline filled with its background color at 20% shows where it will appear. In a progressive reveal set it on one element per step only: the box the step's other elements sit on, see the slide properties). Positioning Elements are not auto-laid-out: left/top/width/height must be computed deliberately against the canvas size, not guessed. Align elements to the slide edges and to each other — matching left edges for a stacked column, centering with left = (canvas width − element width) / 2, equal gaps between siblings — so nothing ends up crowded, overlapping, or visibly off-grid. Labels default to text-align: "center". That suits short titles but is usually wrong for body text, bullet lists, or captions — set text-align (e.g. "left") explicitly whenever centered text would look out of place. Text For labels, buttons, inputs, checks, radio controls, drop-downs, and lists: - text; labels accept HTML and <br> for line breaks - color, font-family, font-size (pixels), font-weight, and font-style - text-align, vertical-align, line-height, and padding.all Size the box to comfortably fit the text, not just match it: estimate the rendered text's width and height from font-size, character count, and line-height (accounting for wrapping), then make width/height larger than that estimate. A box sized exactly to the text leaves no room for padding and risks clipping or crowding it, especially with longer or multi-line text. Set padding.all (commonly 12–24 px, more for larger text) so the text never touches the box edges. Backgrounds For slides, labels, buttons, inputs, images, webviews, charts, and similar elements: - background.type: solid, gradient, or an empty string for none - background.color, background.image, and background.size - background.gradient.type: linear, circle, or ellipse - background.gradient.stops: an array of {color, stop} values - Linear gradient coordinates x1, y1, x2, and y2 use percentages Borders and shadows - Border: border.style.all, border.width.all, border.color.all, and border.radius.all - Shadow: `b…
list_skill_text
ChatGPTRead the complete instructions for one USolver skill. Pass an id or display name returned by list_skills (case-insensitive, with or without the leading slash, for example calculate or /renderScience). Skills may depend on each other (for example calculate, problem and solve all use renderScience). When a skill says to load another skill, call this tool again for it. The text returned is a binding specification for the task, not optional advice. Follow every requirement in it: do not skip, simplify or substitute required elements, controls, properties, defaults or file formats, and do not describe the work as complete until you have checked the result against the skill. If something cannot be done, say which requirement and why.
list_skills
ChatGPTList the available USolver skills that provide instructions for building specific kinds of document content (for example calculators, interactive problems, scientific notation rendering). The list is built dynamically from the MCP skills directory. Each result includes its stable id, display name, and a one-sentence description. A user request that starts with a slash command such as /calculate, /problem, /solve or /renderScience refers to the skill with that id. Always read the skill with list_skill_text before starting the work, and read any other skills it tells you to load. Once a skill applies, it is a binding specification: follow all of its requirements.
list_themes
ChatGPTReturn the slide themes as text for you to review: the built-in themes and the user's saved custom themes ("My themes"), each with its id, name, light or dark, mood, fonts and description. Nothing is shown to the user. Use it whenever you need theme ids, for example the id of a saved custom theme the user named (my-...), or to find the theme closest to a custom theme request. To let the user see and choose themes, use theme_picker instead.
list_user
ChatGPTReturn the user currently authenticated with MCP: user_id, fname, lname, email, doc_count (the number of documents owned by that user, across all folders and visibility settings), doc_max (the document limit, or "unlimited" when zero or unset), and server_url (the server's main public URL). Missing profile fields are returned as null. Takes no parameters. Only returns the authenticated user's profile; it cannot look up other users. This tool does not modify the account or documents.
sheet_read
ChatGPTRead the spreadsheet data of a document. Each slide uses one sheet, its own or one shared with other slides; the slide's sheet is used unless sheet is given. Returns - All sheets and the slides that use them - The cells of the selected sheet as a table (formulas are shown as written)
sheet_update
ChatGPTInsert or change data in the document's spreadsheet (the data behind slides). Use it when data should go into the spreadsheet rather than onto the canvas, and to feed charts and labels connected to cells. Each slide uses one sheet: usually its own, but several slides can share one. The slide's sheet is used unless sheet is given. Writing to a shared sheet changes the data on every slide that uses it. To put data into a new document, create it with doc_create first. Writing data - rows: a table written from start, one array per row. Without start it goes to the first empty rows from column A, so existing data is kept. - cells: single cells with an optional format (mask) and styles, e.g. a bold header or a total formula. - Cell addresses use a1 notation. Formulas start with =, e.g. "=sum(b2:b13)". - Before writing formulas, read the sheet-formulas topic with udoc_ref. The formula language is not Excel: many Excel functions do not exist, and this tool does not report formula errors. - Number-like strings are stored as numbers; "TRUE" / "FALSE" become booleans; "" clears a cell. Returns The written range (for chart properties such as chart.x-data: "a2:a13") and the written cells.
slide_add
ChatGPTAdd a slide to an existing document. The slide may include a background and initial elements. Set kind to pick the theme's slide design: title for the opening slide, section for a slide that starts a new part, content (default) for regular slides and closing for the final summary, thank-you or call to action. The theme gives each kind its own background and decoration; do not set a background unless the user asks for one (an explicit background is kept when the theme changes). The slide gets its own new sheet. To share data with other slides, pass sheet with an existing sheet name (see sheet_read). Slide properties props accepts both per-slide properties and document/deck-wide properties (which apply to every slide, e.g. width, height, canvas-size, navigation.*, page.*) in the same object — each one is automatically applied to the right place (the slide or the document), so there's no need to call them separately. Supported slide properties include: - name - background.type: solid or gradient - background.color, background.image, and background.size - background.gradient.type and background.gradient.stops - border.style.all and border.color.all - feedback: spreadsheet cell that stores the slide's current reveal step - min: minimum value for the feedback cell; defaults to 0 - max: maximum value for the feedback cell; omit it for no upper limit Progressive element reveal To reveal elements progressively, as in a normal presentation, set feedback to a spreadsheet cell and define min and max on the slide. Then use that cell in each element's show-when expression. Each progressively revealed element should also define transition.time as 300 for a 300 ms reveal transition. Mark where each step will appear with show-hidden-outline: true on exactly one element per reveal step: the element that defines the step's box, such as the card, panel or background shape that the step's other elements sit on. While it is hidden, a dashed outline with its exact position and size, filled with its background color at 20%, shows where the step will appear. The outline fades out as the step is revealed (and back in when the step is undone), so the audience sees the layout of the slide from the start. - Do not set it on every hidden element. The other elements of the same step (titles, text, icons or images inside the box) keep it off, otherwise their outlines overlap the box's outline and clutter the slide. - If a step reveals a single element (for example a label with a background), set it on that element. - If no element of a step is a box around the others, set it on the largest one. For example, configure the slide with: ``json { "feedback": "a1", "min": 0, "max": 3 } ` Then assign reveal conditions to elements. Here step 1 is a card (a rectangle) with a title and a text label on it, so only the card gets the outline; steps 2 and 3 are single labels: `json { "name": "Card", "show-when": "a1>=1", "transition.time": 300, "show-hidden-outline": true } { "name": "Card title", "show-when": "a1>=1", "transition.time": 300 } { "name": "Card text", "show-when": "a1>=1", "transition.time": 300 } { "name": "Point 2", "show-when": "a1>=2", "transition.time": 300, "show-hidden-outline": true } { "name": "Point 3", "show-when": "a1>=3", "transition.time": 300, "show-hidden-outline": true } ` During the presentation: - Up Arrow or an unmodified left click increases the feedback cell by 1, revealing the next step, up to max. - Down Arrow or Shift/Ctrl/Command + left click decreases it by 1, hiding the previous step, down to min. - Enter resets the feedback cell to min, returning the slide to its initial reveal state. Use one feedback cell per slide and make sure it belongs to the spreadsheet selected by that slide's spreadsheet property. Looking up all slide properties Use the udoc_ref tool with the props-slide` topic to look up the complete reference - both per-slide and document/deck-wide properties - before using properties that are not listed here:…
slide_delete
ChatGPTDelete a slide and all of its elements from an existing document. Constraint A document must retain at least one slide.
slide_snapshot
ChatGPTReturn a slide snapshot: a generated image preview of the rendered slide, returned as PNG image content for visual inspection. Supply the slide's 0-based index, using a known index or doc_read to identify it if needed. Fix problems and inspect affected snapshots again. Snapshots show the final progressive-reveal state, not every reveal step or interactive state. Availability Slide snapshots are generated when a document is created and regenerated automatically when a slide or its elements are updated and saved. If only one slide changes, reuse the other snapshots rather than requesting a full-deck regeneration. Shared changes can affect multiple slides, and missing snapshots are generated as needed. The tool waits for pending generation and requests missing slide snapshots when necessary. If a snapshot is unavailable, report that the visual review could not be completed.
slide_update
ChatGPTChange an existing slide's properties or move it to another position. Editable properties include the name, background color, gradient, and image. Pass sheet to switch the slide to another existing sheet, so it shares that sheet's data with other slides. Pass kind (title, section, content or closing) to give the slide another of the theme's slide designs: its background and decoration change, elements with a theme role are restyled for that kind (for example light text on a dark section slide). Setting a background in props overrides the theme; such a background is kept when the theme changes. Slide properties props accepts both per-slide properties and document/deck-wide properties (which apply to every slide, e.g. width, height, canvas-size, navigation.*, page.*) in the same object — each one is automatically applied to the right place (the slide or the document), so there's no need to call them separately. Supported slide properties include: - name - background.type: solid or gradient - background.color, background.image, and background.size - background.gradient.type and background.gradient.stops - border.style.all and border.color.all - feedback: spreadsheet cell that stores the slide's current reveal step - min: minimum value for the feedback cell; defaults to 0 - max: maximum value for the feedback cell; omit it for no upper limit Progressive element reveal To reveal elements progressively, as in a normal presentation, set feedback to a spreadsheet cell and define min and max on the slide. Then use that cell in each element's show-when expression. Each progressively revealed element should also define transition.time as 300 for a 300 ms reveal transition. Mark where each step will appear with show-hidden-outline: true on exactly one element per reveal step: the element that defines the step's box, such as the card, panel or background shape that the step's other elements sit on. While it is hidden, a dashed outline with its exact position and size, filled with its background color at 20%, shows where the step will appear. The outline fades out as the step is revealed (and back in when the step is undone), so the audience sees the layout of the slide from the start. - Do not set it on every hidden element. The other elements of the same step (titles, text, icons or images inside the box) keep it off, otherwise their outlines overlap the box's outline and clutter the slide. - If a step reveals a single element (for example a label with a background), set it on that element. - If no element of a step is a box around the others, set it on the largest one. For example, configure the slide with: ``json { "feedback": "a1", "min": 0, "max": 3 } ` Then assign reveal conditions to elements. Here step 1 is a card (a rectangle) with a title and a text label on it, so only the card gets the outline; steps 2 and 3 are single labels: `json { "name": "Card", "show-when": "a1>=1", "transition.time": 300, "show-hidden-outline": true } { "name": "Card title", "show-when": "a1>=1", "transition.time": 300 } { "name": "Card text", "show-when": "a1>=1", "transition.time": 300 } { "name": "Point 2", "show-when": "a1>=2", "transition.time": 300, "show-hidden-outline": true } { "name": "Point 3", "show-when": "a1>=3", "transition.time": 300, "show-hidden-outline": true } ` During the presentation: - Up Arrow or an unmodified left click increases the feedback cell by 1, revealing the next step, up to max. - Down Arrow or Shift/Ctrl/Command + left click decreases it by 1, hiding the previous step, down to min. - Enter resets the feedback cell to min, returning the slide to its initial reveal state. Use one feedback cell per slide and make sure it belongs to the spreadsheet selected by that slide's spreadsheet property. Looking up all slide properties Use the udoc_ref tool with the props-slide topic to look up the complete reference - both per-slide and document/deck-wide properties - before using properties that are not listed here: ``j…
theme_picker
ChatGPTShow the interactive theme picker. Use it only when the user asks to see, browse or choose themes. Do not show it before doc_create just to ask: pick a fitting theme yourself, the document preview lets the user switch themes in one click. To look up themes or ids yourself, use list_themes, which shows nothing to the user. Theme picker - Shows the built-in themes and the user's saved custom themes ("My themes") with rendered sample slides (title, content, data and section slides), filterable by Light / Dark and My themes. - When the user clicks a theme, the choice is sent to the chat as a message. Use that theme id with doc_create, or with doc_theme to restyle an existing document. - The picker also has a Create Theme button: the user is asked to describe the look they want. Then read the custom-theme skill, build the theme JSON, show it with theme_preview, and save it with theme_save once the user confirms. A theme is more than colors: each one defines slide designs (backgrounds and decoration for title, section, content and closing slides) and styles for element roles (titles, cards, callouts, tables, charts, buttons). Pick one whose mood fits the content and audience.
theme_preview
ChatGPTValidate a theme and show it as rendered sample slides (title, content with cards, data with a table and chart, and section). Use it for every custom theme before it is saved, and whenever the user wants to see a theme up close. Custom themes theme is a theme JSON object (or the id of a built-in or saved theme). Read the custom-theme skill before writing one: it describes the format, the available fonts and the design rules. - The theme is validated: invalid colors or fonts and unreadable text (contrast below 3:1) are errors; fix them and call this tool again. Weak contrast (below 4.5:1) is a warning. - Previewing does not save the theme. Show the preview, ask the user whether they like it, and adjust the JSON and preview again until they do. - Only after the user confirms, save it with theme_save (same JSON) and use the my-... id it returns with doc_create or doc_theme.
theme_remove
ChatGPTRemove a custom theme from the user's "My themes". Only saved custom themes (ids starting with my-) can be removed, never built-in ones. Documents that use the theme are not changed: every document keeps its own copy of its theme. Remove a theme only when the user asks for it, and confirm the theme's name with them first.
theme_save
ChatGPTSave a custom theme in the user's "My themes", so it can be used by id with doc_create and doc_theme and shows up in the theme picker and in the USolver Theme button. Call it only after the user has seen the theme in theme_preview and confirmed they like it. Never save a theme the user has not seen, and never save each draft while adjusting it. - theme is the same theme JSON that was previewed (format in the custom-theme skill). It is validated again: fix the errors it reports. - The id is my- plus the theme's name (for example "Ocean Breeze" becomes my-ocean-breeze). Saving a theme with the same name replaces the saved theme; give it a new name to keep both. - Returns the id to pass as theme to doc_create or doc_theme.
udoc_ref
ChatGPTReference documentation for the USolver document (udoc): which element types are supported, documentation for each element type, and documentation for the different property groups used across elements and slides, and what they mean. Usage - List topics: call udoc_ref with no topics to get the full list of available topics, each with a short description. - Read topics: call udoc_ref with topics set to an array of topic names to get their documentation, for example {"topics": ["type-label", "props-text", "props-background"]}. - Element types: every supported element type has its own topic named type-<type>, e.g. type-label, type-button, type-chart-bar. Prefix the element type with type- to get that element's documentation. - Groups: grouping and custom groups are documented by group-* topics. Read group-general first (how the group element and its members work), then the topic of the custom group, for example group-table, group-labeling or group-hotspots. There are other custom groups too, list the topics to see them. - Consult the reference when working with unfamiliar properties, especially charts, quizzes, sliders, events, and gradients. Document properties are the source of truth All document features can be implemented by setting the document's saved properties because those properties are the source of truth. When consulting reference topics, pay particular attention to each property's name, supported element types, value format, dependencies, and meaning. Use that property information through the appropriate MCP mutation tool instead of trying to reproduce the reference's internal scripts. Using in-editor references through MCP The reference documentation supports writing scripts that the internal agent can execute to mutate a document. Do not execute or reproduce those mutation scripts through MCP. All document mutations on the MCP side must be performed through the provided MCP tools. The reference documentation was written for the in-editor agent. On the MCP side there is no app, canvas, or model object. Use these translations. Inspect documents and elements - Documentation: getActiveCanvas(), getCanvas(), canvas.models.get(), model.prop(), and model.rawProp(). - MCP: Use doc_read, then find the required slide or element in its result. It returns saved raw properties; MCP has no active canvas or calculated-property view. Add, update, or remove elements - Documentation: canvas.add(), property-setting model.prop({...}), and internal removal methods. - MCP: Use elem_add, elem_update, and elem_delete. Preserve the documented dot-path property keys. Add, update, move, or remove slides - Documentation: slideAdd(), canvas.prop(), slideMove(), slideRemove(), and slideSwitch(). - MCP: Use slide_add, slide_update, or slide_delete. Use moveTo to reorder and pass slide to the next tool instead of switching the active slide. Slide duplication has no direct shortcut. Read or write spreadsheet cells - Documentation: getCell(), setOriginal(), findEmptyBlock(), and parseChartForAxis(). - MCP: Use sheet_read and sheet_update. Omit start when writing rows to find empty space. Build chart ranges from the returned table. Object-valued cells and internal options such as saveUndo have no MCP equivalent. Manage attachments and previews - Documentation: Internal attachment and rendering APIs. - MCP: Use attach_add, attach_remove, and doc_preview. Look up types and properties - Documentation: findProps(), findTypes(), getTypes(), getAllProperties(), and getProperty(). - MCP: Use the relevant udoc_ref topic. MCP has no runtime property-registry query. Internal-agent helpers - Documentation: agent.history, agent.log, selection, agent.state, agent.msg(), agent.ask(), and agent.error(). - MCP: Do not translate these into tool calls. Use IDs returned by tools and respond to the user directly. Cell-linked…