deckpipe-mcp

MCP server for deckpipe — create, edit, and share slide decks from any AI agent. You describe slides as JSON (layout + content); Deckpipe renders, themes, and exports them. Every deck gets a shareable viewer URL.

Ten tools for the full deck lifecycle: create, read, update, delete, search stock photos, upload images, list layouts, and collaborate via threaded comments. No account or API key required — decks are addressed by unique unguessable IDs.

• Website & viewer: deckpipe.dev
• Remote MCP endpoint: https://deckpipe.dev/mcp (Streamable HTTP)
• npm package: deckpipe-mcp (stdio for local agents)
• Source: github.com/bjoern2000/deckpipe-mcp

Install

Remote (Claude.ai, Claude Desktop, any Streamable HTTP MCP client)

Add https://deckpipe.dev/mcp as a custom remote MCP server. No authentication required.

Local (stdio)

Claude Code

claude mcp add deckpipe -- npx deckpipe-mcp

Manual MCP config (Claude Desktop, Cursor, etc.)

{
  "mcpServers": {
    "deckpipe": {
      "command": "npx",
      "args": ["-y", "deckpipe-mcp"]
    }
  }
}

Tools

| Tool | Description | |------|-------------| | create_deck | Create a new deck and get a shareable viewer URL | | get_deck | Retrieve a deck by ID | | update_deck | Update title, fonts, accent color, slide content, or slide structure (insert, delete, move, replace) | | delete_deck | Delete a deck permanently | | upload_image | Upload a base64 image, get a hosted URL | | search_images | Search Unsplash for stock photos with automatic attribution | | list_layouts | List available layouts, customization options, and style guide | | list_comments | List comments on a deck (filter by status or slide) | | reply_to_comment | Reply to a comment thread after addressing feedback | | resolve_comment | Mark a comment as resolved |

Layouts

| Layout | Description | Required Fields | |--------|-------------|----------------| | title | Large centered title slide | title, optional subtitle, image_url | | title_and_body | Title + paragraph | title, body, optional image_url | | title_and_bullets | Title + bullet list | title, bullets[], optional image_url | | title_and_table | Title + data table | title, table: { headers[], rows[][] } | | two_columns | Title + two columns | title, left: { heading, body }, right: { heading, body } | | section_break | Bold section divider | title | | image_and_text | Image-primary (~60%) + text | title, body, image_url or image_prompt | | image_gallery | Horizontal row of portrait images | images[] (2-5 URLs), optional title, caption | | stats | Big metrics/numbers with labels | metrics[]: { value, label } (2-4), optional title | | quote | Inline curly quotation marks with attribution | quote, optional attribution, image_url | | full_image | Full-bleed background image with overlay text | image_url or image_prompt, optional title, subtitle | | timeline | Continuous timeline with positioned milestones | events[]: { label, title, description?, position? } (3-6). position (0-1) places milestone at relative point; events alternate above/below line | | comparison | Side-by-side comparison with verdict | left: { heading, bullets[] }, right: { heading, bullets[] }, optional title, verdict | | code | Syntax-highlighted code block (18 languages) | code, optional title, language, caption | | callout | Large featured value with context | value, optional title, label, body | | icons_and_text | Icon grid with headings and descriptions | items[]: { icon, heading, description? } (3-6), optional title. icon accepts any Lucide icon name (e.g. "clock", "message-square") or emoji | | team | Team member cards with roles and bios | members[]: { name, role, bio?, image_url? } (1-6), optional title | | embed | Full-slide iframe embed (90% area) | url, optional caption, aspect_ratio | | pros_and_cons | Two-column pros vs cons list | pros[], cons[], optional title, pros_heading, cons_heading | | agenda | Numbered agenda items with durations | items[]: { topic, duration?, description? } (1-10), optional title | | swot | Four-quadrant SWOT analysis with emoji headers | strengths[], weaknesses[], opportunities[], threats[] (1-5 each), optional title | | quadrant | 2D scatter plot with labeled axes and items | items[]: { label, x: 0-1, y: 0-1 } (1-12), optional title, body, bullets[], x_label, y_label, quadrant_labels[4] | | venn_diagram | 2 or 3 overlapping circles with labels and items | circles[]: { label, items[]? } (2-3), optional title, body, overlaps[]: { sets[], label } | | closing | Accent-colored ending slide with contact info | optional heading, subheading, contact_lines[], image_url | | chart | Bar, line, pie, or donut chart from structured data | chart_type ("bar"/"line"/"pie"/"donut"), data: { labels[], datasets[]: { label?, values[], color? } }, optional title |

All layouts support an optional key_takeaway field — a highlighted sentence rendered below the title.

See all 25 layouts in action →

Image placeholders

Any layout that accepts image_url also accepts image_prompt as an alternative. When set, a dashed placeholder box is rendered showing the prompt text — useful when you want to create a complete deck without sourcing images, letting the user drop them in later.

{
  "layout": "title_and_bullets",
  "content": {
    "title": "Electric Vehicles",
    "bullets": ["Range anxiety solved", "Charging network expansion"],
    "image_prompt": "Hero shot of a sleek EV on a mountain road at golden hour"
  }
}

Stock photo search

Use the search_images tool to find stock photos from Unsplash. Results return simple IDs and thumbnails — use the ID as image_ref in your slides and Deckpipe handles attribution, URLs, and download tracking automatically.

// Search returns IDs + thumbnails
{ "results": [{ "id": "uimg_abc123", "thumb": "https://...", "alt": "mountain landscape" }] }

// Use image_ref in slides — no attribution needed
{
  "layout": "full_image",
  "content": {
    "image_ref": "uimg_abc123",
    "title": "Mountain Adventure"
  }
}

Use queries to search for multiple terms in one call (max 5) instead of making separate calls per slide:

{ "queries": ["sunset beach", "modern office", "mountain hiking"], "orientation": "landscape" }

• image_ref resolves to a real Unsplash URL with proper attribution caption
• For image_gallery, use image_refs (array of IDs) instead of images
• Use orientation: "landscape" for full_image/image_and_text, "portrait" for image_gallery

Rich bullets

Any bullets[], pros[], cons[], strengths[], weaknesses[], opportunities[], or threats[] field accepts either plain strings or rich bullet objects:

{
  "text": "Model 3 leads in range",
  "detail": "EPA-rated 358 miles; closest competitor is 270 miles",
  "sources": [
    { "label": "EPA 2024", "url": "https://fueleconomy.gov" }
  ]
}

• detail renders as a hover tooltip (ℹ icon) — great for adding nuance without cluttering the slide
• sources render as superscript numbers with a footnote row at the bottom of the slide

Plain strings still work as before (backward compatible).

Slide operations

update_deck supports structural slide changes via the slide_operations field — an ordered array of operations executed sequentially before any content edits.

| Operation | Fields | Effect | |-----------|--------|--------| | delete | index | Remove slide at index | | insert | index, slide | Insert new slide at index (shifts others down) | | move | from, to | Move slide from one position to another | | replace | index, slide | Replace slide entirely (new layout + content) |

Operations are applied in order, so each operation sees the array as modified by previous ones. Content edits in slides use indices relative to the post-operations array.

{
  "deck_id": "dk_abc",
  "slide_operations": [
    { "op": "delete", "index": 3 },
    { "op": "insert", "index": 1, "slide": { "layout": "title_and_body", "content": { "title": "New Slide", "body": "Hello" } } },
    { "op": "move", "from": 0, "to": 4 }
  ],
  "slides": [
    { "index": 2, "content": { "title": "Updated Title" } }
  ]
}

Customization

Both create_deck and update_deck accept optional styling fields:

| Field | Description | Default | |-------|-------------|---------| | heading_font | Google Font for headings (e.g. "Playfair Display", "Space Grotesk") | DM Sans | | body_font | Google Font for body text (e.g. "Inter", "Roboto") | DM Sans | | accent_color | Hex color (e.g. "#ff6600") | #7c3aed (purple) |

Example prompts

Create a deck about the rise of electric vehicles. Search the web for relevant photos and use them. Use heading font "Space Grotesk", body font "Inter", and accent color "#0ea5e9".

Create a deck about the Apollo 11 moon landing. Search the web for historic NASA photos and use them as full-bleed images. Include a timeline of the mission, stats on the Saturn V rocket, and a famous quote from Neil Armstrong. Use heading font "Libre Baskerville" and accent color "#1e3a5f".

Build a lecture deck on how neural networks learn. Use a comparison table for activation functions, bullet points for key concepts, and add a key takeaway to each slide.

Create a travel guide deck for Tokyo. Search the web for photos and include an image gallery of must-visit spots, stats on tourism, and a full-bleed image cover slide.

Warnings

Both create_deck and update_deck responses may include a warnings array with actionable feedback for the agent:

{
  "deck_id": "dk_abc",
  "viewer_url": "...",
  "warnings": [
    "Slide 0 (title): unrecognized content field \"body\" — this field was ignored. Valid fields: image_focus, image_prompt, image_url, key_takeaway, subtitle, title",
    "Slide 2: image_url returned HTTP 404 — image may not render (https://example.com/missing.jpg)"
  ]
}

Unrecognized content fields — If you pass a field name that doesn't exist for that layout (e.g. body on a title slide), the field is silently dropped but a warning tells you exactly what happened and lists valid fields for that layout.

Unreachable image URLs — All image_url values are checked with a HEAD request during creation/update. If a URL returns an error or is unreachable, a warning is returned so you can fix it with a follow-up update_deck call.

Comments

Deckpipe supports threaded comments for collaborative feedback between users and agents. Users place comments on specific slide elements in the viewer; agents read and respond to them.

Workflow for agents:

1. Call get_deck — each slide includes a comments[] array with open comments
2. Each comment has a content_path (e.g. "title", "bullets[2]", "slide") telling you which field it refers to
3. Use update_deck to address the feedback
4. Call reply_to_comment to explain what you changed
5. The user resolves the comment once satisfied

Agent identity: Set agent_name when calling create_deck (e.g. "Acme Strategy Agent") — this name appears on your replies.

Configuration

By default the MCP server connects to https://deckpipe.dev. To use a different API endpoint:

{
  "mcpServers": {
    "deckpipe": {
      "command": "npx",
      "args": ["-y", "deckpipe-mcp"],
      "env": {
        "DECKPIPE_API_URL": "http://localhost:3010"
      }
    }
  }
}

Support

• Questions, bug reports, feature requests: open an issue on GitHub
• Email: [email protected]

Legal

• Privacy policy — what data is stored, retention, GDPR rights
• Terms of use — acceptable use, liability

Note: Decks are stored without authentication and are accessible by anyone who has the deck ID. Do not store confidential or personal information in decks.

License

MIT © Björn Schefzyk