KitchenSink4PPT

KitchenSink4PPT

Build and revise PowerPoint decks with your AI assistant, with diagrams made of shapes you can still edit.

Your AI gets 193 actions in PowerPoint, a backup before every change, and a question instead of a guess when your request could mean more than one shape.

KitchenSink4PPT is an MCP server that lets Claude Code, the Codex CLI, Copilot CLI, Claude Desktop and other local MCP clients create and edit .pptx files. Its diagrams are built from native PowerPoint shapes and glued connectors, so the slide is still yours to adjust after the AI is done. This is the free Community edition, licensed under the AGPL-3.0.

Product page · llms.txt · Changelog

Platform: the file tools run on Windows and Linux (macOS is untested in this release). Editing a deck that is open in PowerPoint needs Windows with PowerPoint. PDF and image export need PowerPoint on Windows or a supported LibreOffice.

Install

The package needs Python 3.12 or newer. Run the commands below in PowerShell or a terminal, not in an AI chat.

Claude Code or Codex CLI. Install uv, run the line for your app, and start a new session:

claude mcp add ppt -s user -- uvx kitchensink4ppt

codex mcp add ppt -- uvx kitchensink4ppt

Claude Desktop. Install uv and restart Claude Desktop. Download the .mcpb file from the latest release, then open Settings > Extensions > Advanced settings > Install extension and choose the file. The first launch downloads the Python package, so it needs a network connection.

Any other local MCP client. Use uvx as the command and kitchensink4ppt as its argument, or install the package and run kitchensink4ppt as the server command:

pip install kitchensink4ppt

Optional: pip install "kitchensink4ppt[metrics]" adds measurement with your installed fonts for text fitting and overflow checks. The results are still estimates, so render the slide to confirm.

What it can do

• Build a deck from a template and revise its text and speaker notes.
• Draw diagrams from editable shapes and glued connectors, or generate timelines, org charts, matrices, cycles and comparisons in one call.
• Turn SVG graphics into native PowerPoint shapes.
• Align objects and apply consistent fonts, colors, themes and layouts.
• Build tables and charts that colleagues can keep editing.
• Merge, split and compare decks, and work through review comments.
• Check slides for accessibility issues and apply the supported repairs.
• Export PDF or slide images when a rendering application is installed.

In total the server offers 145 operations in 9 tools, 193 counting each action an operation can take (for example each kind of transition or alignment). The whole tool list costs about 7.3k tokens of your AI's working memory at startup. Figures come from scripts/measure_surface.py and scripts/count_operations.py.

KitchenSink4Word (Word), KitchenSink4XL (Excel) and KitchenSink4Web (web pages) are built the same way.

Real graphics, not pictures of graphics

svg_to_shapes compiles an SVG into a grouped tree of native PowerPoint shapes: custom geometry, gradients, strokes, text and nested groups. Nothing is rasterized, no external service is called, and building does not need PowerPoint installed. Every shape keeps its own id, so you can recolor one node, resize a box or rewrite a label afterward.

Connectors are glued. Move a node and its arrows follow, exactly as if someone had drawn the diagram by hand, so the person you hand the deck to can adjust any piece without regenerating it.

insert_shape (presets and freeform geometry), insert_connector (straight, elbow, curved, with arrowheads), group_shapes, align_shapes, distribute_shapes and set_z_order cover diagrams built from scratch, and export_slide_image renders a PNG so your AI can look at what it made and fix it.

Tool packs and context cost

The server lists nine tools, all from the first message. Every tool except ppt_help takes an op (the operation to run) and args (its arguments, including the path to the .pptx). Everything is available from the start; nothing needs switching on. Token figures count the full tools/list entry the client receives and come from scripts/measure_surface.py. ppt_help returns the arguments and a working example for any operation, and unknown arguments are refused rather than ignored. Name a slide with slide_number, which counts from 1 the way people do, or with slide, which counts from 0; every answer about one slide gives its number and title.

For a recipe, ask your AI to call get_workflows. It includes the build-a-diagram, template and table sequences with the right tool for each step. The server's own instructions add about 500 tokens to the figures above. Administrators who want a narrower server can set KS4P_READ_ONLY=1 or list what to remove in KS4P_FORBID (see Configuration); what is removed no longer counts toward the cost.

Safety model

• Atomic, validated saves. Every change rebuilds the package in memory, validates it, then swaps the new file in for the old one. A failed operation leaves the original exactly as it was. On Windows the deck keeps its Mark of the Web, so a downloaded deck still opens in Protected View, and its creation time; it keeps its own permissions wherever the file system and the folder allow it; where they do not, the deck takes its folder's permissions and the answer says so in a warning. Other programs may see the deck missing for a moment during a save, usually a few milliseconds; this server's reads wait it out.
• Backups before every change. The current version rotates into prev.pptx and anchor.pptx in a hidden .ks4p-backups/ folder next to the deck. create_snapshot keeps a permanent copy in the same hidden folder, where nothing rotates or purges it; with dest_dir it writes a visible copy there instead, named after the deck so it never carries a newer leading date stamp than the deck. manage_backups lists and restores them all (a restore can itself be undone) and purges the rotating ones.
• Untouched parts stay untouched. Parts a call does not change are written back byte for byte, so themes, media and animations survive edits to other slides.
• Refusals instead of guesses. Ambiguous targets are refused with a list of candidates, merged-cell edits that would split a span are refused, and a batch with a stale anchor is refused before anything changes. Each refusal is structured ({ok: false, error: {code, message, hint}}) and names the next call to make.
• Locks. Changes to one file are serialized within a server and across processes. A deck open in PowerPoint is read from its saved file, and the answer says so; the file tools refuse to change it rather than risk it.
• PowerPoint stays yours. The server ends a PowerPoint process only if it started that process itself, and it cancels a hung call safely instead of killing the application.
• Encrypted and read-only files. An encrypted presentation is refused before any Office call, and a read-only file is reported as read-only.

Privacy

The server runs on your computer. KitchenSink4AI never receives your files or usage data. Your AI app may send prompts, file contents and tool results to its own provider under that app's settings and terms. The update check described below contacts PyPI at most once a week and sends only the package name and the installed version. Choosing Update now downloads the new version from PyPI, or the new bundle from GitHub for Claude Desktop. The server's startup banner is switched off, so the framework it runs on makes no version check of its own. The Privacy Policy covers the product, purchases and support.

Configuration

| KS4P_AUTHOR | Who signs a comment or reply when the call names no author: user (the default) s