PowerPointMcp — MCP Server & CLI for Microsoft PowerPoint

Automate PowerPoint with AI — a Model Context Protocol server and CLI for live, real-time PowerPoint automation through conversational AI. Sibling project to mcp-server-excel, following the same layered architecture: ComInterop → Core → CLI / MCP Server.

MCP Server for PowerPoint enables AI assistants (GitHub Copilot, Claude, ChatGPT) to build and edit real .pptx presentations through natural language — slides, shapes, text, tables, charts, speaker notes, and layouts — no VBA or PowerPoint object-model knowledge required.

🛡️ Live COM automation, not file parsing — Most PowerPoint MCP servers manipulate .pptx files offline with libraries like python-pptx, or use agent-run scripts with LibreOffice-rendered thumbnails. This project instead drives a live, real PowerPoint desktop instance via Microsoft.Office.Interop.PowerPoint — the official Primary Interop Assembly. That means true-fidelity rendering, compatibility with an already-open deck, and zero risk of producing a .pptx that PowerPoint itself can't open, because PowerPoint is the one writing it.

🖼️ Export-to-verify — the core differentiator. After any visual edit, export the slide (or the whole deck) to an image with export(action="export-slide-to-image", ...) / export(action="export-all-slides-to-images", ...) and let a vision-capable AI assistant see the result — catching overlapping shapes, text overflow, and layout regressions that text-only automation simply cannot detect.

Technical Requirements:

• ⚠️ Windows Only — COM interop is Windows-specific
• ⚠️ PowerPoint Required — Microsoft PowerPoint 2016 or later must be installed
• ⚠️ Desktop Environment — controls a real PowerPoint process (not for server-side processing)

[!TIP] Also automating spreadsheets? Check out Excel MCP Server — the sister project, built the same way.

🎯 What You Can Do

13 MCP tools with 137 operations across 13 domains:

• 🗂️ Presentation (12 ops) — create, open, save, close, list sessions, apply a .potx/.pptx template's masters/theme/layouts, read the current theme name, read/write built-in and custom document properties
• 📑 Slide (14 ops) — add, count, delete, duplicate, reorder, per-slide background color, sections
• ▭ Shape (36 ops) — rectangles, text boxes, auto shapes, lines, connectors, fill/line/shadow, rotation, flip, z-order, grouping, naming, alt text
• ✏️ TextFrame (20 ops) — text, font size/name/color, bold, italic, underline, alignment, bullets
• 📊 Table (12 ops) — add, cell text, insert/delete rows & columns, cell fill/border, merge cells
• 🗣️ Notes (2 ops) — set/get speaker notes
• 🖼️ Layout (2 ops) — set/get slide layout
• 🎭 Master (8 ops) — slide master title/body placeholder fonts, background color
• 🎬 Animation (5 ops) — shape entrance/emphasis/exit effects, slide transitions
• 🖼️ Image (7 ops) — insert and adjust pictures (brightness, contrast, recolor, crop)
• 📈 Chart (10 ops) — add chart, multi-series data, titles, axis titles, legend
• 🔀 SmartArt (7 ops) — insert and edit SmartArt diagrams
• 🖼️ Export (2 ops) — export a slide, or all slides, to images for visual verification

Every domain is exposed as a single action-dispatch tool (e.g. shape, table, chart, presentation) with an action parameter selecting the specific operation — keeping the tool list small for AI assistants while still exposing every operation.

📚 Complete Feature Reference → — detailed documentation of every tool and operation

💬 Example Prompts

Build a deck from scratch:

• "Create a new presentation with a title slide and three content slides about our Q3 results, then export it as images so I can see it."

Tables & charts:

• "Add a 4x3 table summarizing this data, then add a bar chart next to it."

Formatting & shapes:

• "Make the title bold and blue, and move the logo to the top-right corner."

Speaker notes:

• "Write speaker notes for each slide summarizing the key talking point."

Templates & themes:

• "Apply our corporate template to this deck without losing any of the slide content."

Visual verification:

• "Export slide 3 as an image and tell me if the chart overlaps the text box."

👥 Who Should Use This?

Perfect for:

• ✅ AI assistants and coding agents that need to build or edit .pptx decks
• ✅ Anyone automating repetitive slide-deck workflows (reports, status decks, templates)
• ✅ Teams that want export-to-verify visual checks on every automated edit

Not suitable for:

• ❌ Server-side/headless processing (this drives a real desktop PowerPoint process)
• ❌ Linux/macOS users (Windows + PowerPoint installation required)

🚀 Quick Start

⚠️ Important: Close any open instances of the target file before automating it — the server needs exclusive access to the presentation while it's driving PowerPoint.

🔧 CLI vs MCP Server

This project provides both a CLI and an MCP Server interface. Choose based on your use case:

Manual installation:

# Primary: Download standalone executables from latest release (no .NET runtime required)
# https://github.com/sbroenne/mcp-server-powerpoint/releases/latest
# - PowerPointMcp-MCP-Server-{version}-windows.zip → extract mcp-powerpoint.exe
# - PowerPointMcp-CLI-{version}-windows.zip → extract powerpointcli.exe

# Secondary: Install via .NET tool (requires .NET 10 runtime)
dotnet tool install --global Sbroenne.PowerPointMcp.McpServer
dotnet tool install --global Sbroenne.PowerPointMcp.CLI

# After installing either way, auto-configure your coding agents:
npx add-mcp "mcp-powerpoint" --name powerpoint-mcp

# Optional: Install the agent skill for better AI guidance
npx skills add sbroenne/mcp-server-powerpoint --skill powerpoint-mcp

💡 The VS Code extension installs this skill automatically. Manual npx skills add is for other MCP clients (Claude Code, Cursor, Windsurf, etc.).

⚙️ How It Works - COM Automation & Unified Service Architecture

PowerPointMcp uses Windows COM automation to control the actual PowerPoint application (not just .pptx files).

The MCP Server and CLI are two equal, first-class entry points. Each hosts its own PowerPointMcp Service that manages presentation sessions — the MCP Server runs it in-process (direct calls, no pipe), while the CLI uses a background daemon over a named pipe so sessions persist across CLI invocations:

┌──────────────────────┐        ┌──────────────────────┐
│  MCP Server          │        │  CLI (pptcli)        │
│  (AI assistants)     │        │  (coding agents)     │
└──────────┬───────────┘        └──────────┬───────────┘
           │ in-process                     │ named pipe →
           │ (direct calls)                 │ background daemon
           ▼                                ▼
┌──────────────────────┐        ┌──────────────────────┐
│  PowerPointMcp       │        │  PowerPointMcp       │
│  Service             │        │  Service             │
│  (session mgmt)      │        │  (daemon; sessions   │
│                      │        │   persist across     │
│                      │        │   CLI invocations)   │
└──────────┬───────────┘        └──────────┬───────────┘
           ▼                                ▼
      Core Commands                    Core Commands
           ▼                                ▼
┌──────────────────────┐        ┌──────────────────────┐
│  PowerPoint COM API  │        │  PowerPoint COM API  │
│  (PowerPoint.        │        │  (PowerPoint.        │
│   Application)       │        │   Application)       │
└──────────────────────┘        └──────────────────────┘

Both entry points share the same Core Commands codebase, so every operation behaves identically. They are separate processes, though: each runs its own PowerPointMcp Service and its own PowerPoint instance, and they do not share live sessions with each other.

Key Benefits:

• ✅ Two equal entry points — every operation works identically through the MCP Server and the CLI
• ✅ Persistent CLI sessions — the CLI daemon keeps presentations open across multiple pptcli calls, so scripts don't re-open files each time
• ✅ In-process MCP calls — the MCP Server runs the service in-process (no pipe) for low-latency automation
• ✅ Real PowerPoint automation — drives the actual PowerPoint.Application via COM, not just file parsing
• ✅ Export-to-verify — close the loop on every visual change with a real rendered image

⭐ GitHub Star History

Updated daily from GitHub's stargazer data.

📋 Additional Information

📚 MCP Server Guide → | CLI Guide → | Agent Skills →

📖 Complete Feature Reference • Installation Guide • Changelog • Contributing • Security • Privacy

License: MIT License — see LICENSE

Built With: This entire project was developed using GitHub Copilot AI assistance.

Acknowledgments:

• Microsoft PowerPoint Team — for comprehensive COM automation APIs
• Model Context Protocol community — for the AI integration standard
• mcp-server-excel — the sibling project this architecture and tooling is ported from

Related Projects

Other projects by the author:

• Excel MCP Server — AI-powered Excel automation via Power Query, DAX, VBA, and PivotTables
• Windows MCP Server — AI-powered Windows automation via MCP
• OBS Studio MCP Server — AI-powered OBS Studio automation
• pytest-skill-engineering — LLM-powered testing framework for AI agents