OfficeAgent.NET: .NET Office Automation for AI Agents

What changed

OfficeAgent.NET is a .NET library and MCP server designed for structured automation of Microsoft Word, PowerPoint, and Excel files. It enables AI agents and .NET applications to edit these documents without breaking their internal structure, preserving styles, comments, and other elements. The system operates using the Microsoft Open XML SDK and does not require Office to be installed on the machine.

Key capabilities include:

• Word Automation: Filling content controls, bookmarks, and repeating table rows in Word templates while maintaining original styles, numbering, and comments. It can also record edits as tracked changes that Word can accept or reject, and perform document comparisons.
• PowerPoint Automation: Building new slides based on existing template layouts, populating named shapes, tables, charts, and pictures, while leaving logos and fonts untouched.
• Excel Automation: Inspecting and editing worksheets, tables, and bounded ranges, including reading cell values as displayed, setting cells and formulas, appending table rows, and managing cell notes.

The engine is available in three forms: an MCP server for agents supporting the Model Context Protocol, tools for Microsoft Agent Framework and Microsoft.Extensions.AI, and a direct .NET API for applications. It supports .docx, .pptx, and .xlsx file formats, routing each document to the appropriate module.

Recent updates include version v1.3.0, which offers standalone downloads for running the MCP server without .NET installation and integration options for Claude Desktop. The project also provides samples for Python integration with OpenAI Agents SDK and LangGraph, and a contract review skill for Microsoft 365 Copilot.

Why it matters for builders

This project offers AI builders a robust solution for programmatic document manipulation, bridging the gap between AI capabilities and the ubiquitous Microsoft Office file formats. Developers can integrate these tools into agent frameworks like Claude Code, Cursor, or custom .NET applications, enabling AI to perform complex tasks such as generating reports from data, updating presentations, or processing contractual documents with precision.

The ability to work with Office files without requiring Office installation significantly reduces deployment complexity and opens up possibilities for server-side automation and cloud-based agent solutions. The structured approach, using typed plans and previews, ensures that operations are validated before execution, minimizing errors and data corruption.

Practical impact

Developers can start by exploring the provided samples, such as the "Try a Word edit" workflow, which demonstrates changing text within a document and applying it as a tracked change. Installation of the MCP server version 1.3.0 is straightforward using dotnet tool install --global OfficeAgent.Mcp --version 1.3.0. The project offers clear guidance on how to integrate OfficeAgent.NET with various AI clients, including Claude Code, Cursor, and Microsoft Copilot, via MCP over stdio or HTTP.

For .NET developers, the dotnet add package OfficeAgent.Word command initiates integration into existing applications. The project also details how to connect to document storage services like SharePoint, Azure Blob Storage, and Amazon S3, expanding its utility for enterprise scenarios. Builders looking to automate report refreshes or perform bulk edits across multiple documents will find specific workflows and samples tailored to these needs.

Caveats and source limits

The source material indicates that while OfficeAgent.NET handles many aspects of document structure, it may have limitations regarding Office's specific layout or calculation engine. Users should consult the "Scope and limitations" section for detailed information on these potential constraints. The project is presented as a fresh release with 28 stars and 8 forks on GitHub, suggesting it is still gaining community traction. Specific performance benchmarks or independent validation of its capabilities against complex, real-world documents are not detailed in the provided excerpts.