Work with saved Web Builder decks through MCP

Deckary's saved-deck MCP tools let an external AI agent work with presentations stored in your Deckary Web Builder account. PowerPoint does not need to be open. Access is owner-scoped, and each authoring request is checked against the current deck revision before Deckary commits it.

Discover a saved deck#

Call list_decks to retrieve the signed-in user's saved presentations. Results include stable deck IDs, current revisions, slide counts, and update times. The optional query is limited to 100 characters, and the result limit can be 1–100 with a default of 25.

An agent should always discover the deck instead of guessing a deckId.

Read only the slides you need#

Call read_deck with the chosen deckId. Without slideIds, the tool returns compact deck state and the first slide. With up to ten current slide IDs, it returns the current revision, outline, theme, and exact editable scene truth for those slides.

The returned revision is the concurrency boundary for later authoring. If another edit changes the deck, the agent must read the newer state before preparing a replacement program.

Render slide images#

render_slides accepts one to four current slide IDs and returns exact 960×540 PNG images. Each image is bound to the current deck revision and an artifact fingerprint.

Use renders to inspect the deck before a material edit and to evaluate the visual result returned by authoring. A stale image should not be treated as evidence of the current deck.

Author against the current revision#

author_deck accepts the deckId and a semantic Deckary authoring program. The program can create, revise, reorder, or remove supported presentation content. Deckary validates the complete program and applies it against the expected revision.

The author_deck description lists the available operations and safety rules only. Before the first author_deck call, have the agent call describe_authoring_vocabulary with topic web_deck_scene for the complete scene, node, text, chart, table, media, and patch vocabulary, and again whenever a validation issue reports an unknown key. A successful result returns current deck state and up to four images for committed changed slides. A stale revision is rejected rather than overwriting newer work. The external agent should use that returned state for its next decision instead of immediately repeating discovery calls.

Add a consulting visual#

Consulting visuals such as icebergs, funnels, staircases, issue trees, and strategy houses are nodes in the same author_deck program. There is no separate preparation step. A consulting_visual node names a catalog recipe and gives its content in the recipe's own terms:

{
  "id": "maturity",
  "type": "consulting_visual",
  "placement": { "mode": "absolute", "frame": [72, 120, 816, 360] },
  "recipe": "deckary_straight_staircase",
  "content": { "steps": ["Ad hoc", "Defined", "Managed", "Predictive"] },
  "color": "accent1"
}

Call describe_authoring_vocabulary with topic consulting_visuals for when a visual is the right exhibit and for every recipe's content shape. Deckary derives the layout from the content and fits it inside the frame, and renders any artwork the recipe needs as a reusable deck asset during the same call. Content that does not match the recipe, such as an iceberg with two drivers instead of three, is rejected with an issue at its exact path and changes nothing. A successful result may include advisory notes, for example when copy will shrink to fit.

read_deck returns an untouched visual in the same recipe and content form, so the agent can revise it by upserting the node. If someone has edited the visual's shapes by hand, it reads as an ordinary group instead, and the slide lists it under editedConsultingVisuals with its original recipe and content.

Recommended saved-deck sequence#

1. list_decks to select the saved presentation.
2. read_deck for compact state and only the target slides.
3. describe_authoring_vocabulary (topic web_deck_scene) before the first authoring call.
4. describe_authoring_vocabulary (topic consulting_visuals) when the edit includes a consulting visual.
5. render_slides if visual context is required.
6. author_deck with the current revision and complete semantic program.
7. Inspect the committed images, current state, and any notes returned by authoring.

See the tool reference for the complete saved and live tool set, or Getting started for connection instructions.