> ## Documentation Index
> Fetch the complete documentation index at: https://docs.prisme.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Slides & Deck Designer
> Turn a brief into a professional, on-brand slide deck — live preview in the canvas, editable PPTX + PDF export, all client-side
**Deck Designer** turns a brief into a professional, on-brand slide deck. The agent writes a *deck spec*, the [canvas](/products/ai-securechat/canvas) renders it as a live preview, and the same render exports to an **editable PPTX** and a **pixel-perfect PDF** — entirely client-side, with no backend and no data leaving your infrastructure.
It is built from three composable pieces — a **skill**, an **MCP server**, and the **canvas render/export engine** — so you can attach the same capability to any agent (sales, marketing, support…), not just a dedicated "Slides" agent.
## The model in one line
**One spec → one render → two exports.** The agent produces a *deck spec* (JSON). A theme-driven renderer turns it into a live HTML preview, and the same render is exported client-side to an editable PPTX (mapped from the rendered DOM geometry) and a pixel-perfect PDF. Look & feels are **data** (theme tokens served by an MCP server), so a new style is a new data entry — no code, no agent change.
## How it works
```mermaid theme={null}
flowchart LR
A[Agent + deck-designer skill] -- "get_theme(name)" --> B[Slides MCP - theme registry]
B -- "theme tokens (colors, fonts)" --> A
A -- "deck spec JSON in pr-canvas artifact" --> C[Canvas]
C -- "live HTML preview" --> D[User]
C -- "PPTX (editable) and PDF" --> E[Download]
```
1. The user asks for a deck / slides / pitch / `.pptx`. The agent activates the **deck-designer** skill (progressive disclosure — see [Capabilities](./capabilities)).
2. The skill resolves a look & feel by calling the **Slides MCP** (`list_themes` → pick a style, `get_theme(name)` → tokens).
3. The agent writes the **deck spec** (slides + resolved theme tokens) and emits it as a canvas artifact: `{"type":"slides"}--- `.
4. The **canvas** renders the spec to a live HTML preview and offers **PPTX** (editable) and **PDF** exports — both run in the browser.
Export is layout-agnostic. The PPTX is built from the **rendered DOM geometry** (`getBoundingClientRect` + `getComputedStyle`) mapped to native PowerPoint shapes/text — so any slide the renderer emits (including freeform HTML slides) exports to an editable `.pptx` with no per-layout code. The PDF rasterizes each rendered slide (landscape 16:9, one page per slide).
## The three pieces
| Piece | What it is | Where it lives |
| ------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------ | ----------------------------------------------------------------------------------- |
| **`deck-designer` skill** | Owns the deck-spec schema, the design system (palettes, typography, layout catalogue, anti-patterns), the QA discipline, and how to emit the canvas artifact | Published in the shared **capability Store** (the `capabilities` workspace catalog) |
| **Slides MCP** | A theme registry — `list_themes()` and `get_theme(name)` return look & feel token sets (colors, fonts) | A lightweight Prisme.ai workspace (`slides-mcp`) |
| **Canvas render/export engine** | Renders the deck spec to the preview and runs the PPTX/PDF export | Built into the chat client (SecureChat canvas) |
## Add Deck Designer to an agent
You bring deck building into any agent in **three steps** — open the agent and go to **Capabilities** (see [Agent Capabilities](./capabilities)):
**Add Capability → Skills → Deck Designer.** The skill carries its own instructions (the design system) and is loaded on demand when the user's request matches.
**Add Capability → MCP server**, pointing at the Slides MCP endpoint (name it e.g. `slides-themes`):
```
https:///v2/workspaces/slug:slides-mcp/webhooks/mcp
```
This is what `get_theme` / `list_themes` call to resolve a look & feel.
Turn on **`canvas_enabled`** for the agent. Without it, the agent never learns to emit the deck artifact, so the preview and export won't appear.
All three are required. The skill alone resolves no theme and produces no preview; the MCP alone is never called; and without `canvas_enabled` the deck artifact is never emitted. A multi-skill agent keeps all its own tools — the model activates `deck-designer` only when the user asks for a deck.
## Themes (look & feel)
Themes ship as **data** in the Slides MCP, so re-skinning a deck is just picking another style — the content is untouched.
| Theme | Style |
| ----------------------- | -------------------------------------------------------------------- |
| `executive` *(default)* | Formal — navy + gold, serif headlines. Finance / consulting / board. |
| `modern` | Clean tech-SaaS — white + teal, all-sans, airy whitespace. |
| `bold` | Energetic startup-marketing — high contrast, warm coral accent. |
Ask the agent *"what styles are available?"* (`list_themes`) or *"make the same deck in modern"* (`get_theme("modern")`) — the theme tokens are injected into the spec without rewriting the content.
## Customizing the skill and the MCP server
Both pieces are meant to be tailored to your brand and use cases.
### Add or edit a theme (customize the MCP server)
A new brand or style is a **new data entry** — no code, no agent change:
1. Open the `slides-mcp` workspace and edit `config.value.themes` (in `index.yml`).
2. Add an entry with your palette and fonts. Colors are **bare hex** (no leading `#` — the renderer adds it). Use cross-platform fonts (Georgia / Helvetica / Arial / Trebuchet MS) so the browser preview and the exported PPTX render with matching metrics.
```yaml theme={null}
config:
value:
themes:
my-brand:
name: My Brand
dark: 0B2540 # bare hex, no leading #
light: F4F9FC
ink: 16283C
muted: 5B6B7C
accent: 009DE0
soft: D6F0FB
line: DFE9F1
fontHead: Arial
fontBody: Arial
headWeight: 700
```
`list_themes` exposes the new style automatically — the agent can offer it with no further change. You can also self-host your own copy of the MCP workspace and point your agent's MCP capability at it.
### Customize the skill instructions
The `deck-designer` skill lives in the shared **capability Store** (the `capabilities` workspace catalog, exposed via the `v1/servers` API). A skill entry's instructions live in its `config_schema` defaults, so you can:
* **Fork it for your team** — publish your own `type: skill` entry (e.g. tighter brand rules, your preferred layouts, a fixed default theme) and add that instead.
* **Use the Custom Skill template** — the catalog ships a generic **Custom Skill** entry: create a brand-new skill from scratch with your own instructions, activated on demand via progressive disclosure.
Keep the skill description sharp — it's the trigger the model sees, so it should clearly say *"decks / slides / presentation / pitch / .pptx"*. And prefer a capable model on the host agent: the design discipline is demanding and drives deck quality.
## Export formats
Exports run entirely in the browser — what you see in the preview is exactly what exports.
Native PowerPoint shapes and text boxes, mapped from the rendered geometry — fully editable in PowerPoint / Google Slides / Keynote.
Pixel-perfect, landscape 16:9, one page per slide — faithful to the preview.
## Next Steps
How skills and MCP servers attach to an agent
The collaborative workspace that renders and exports the deck
Wire your own Builder automations as agent tools