Sbroenne.ExcelMcp.McpServer 2.4.1

ExcelMcp - Model Context Protocol Server for Excel

mcp-name: io.github.sbroenne/mcp-server-excel

Control Excel with Natural Language through AI assistants like GitHub Copilot, Claude, and ChatGPT. This MCP server enables AI-powered Excel automation for Power Query, DAX measures, VBA macros, PivotTables, Charts, and more.

➡️ Learn more and see examples

⚡ Powered by the Real Excel Engine

Unlike file-parser libraries that rewrite .xlsx files directly, ExcelMcp drives the actual Excel application through its official COM API. That means it can run live operations file-based tools can't — refresh Power Query, recalculate, refresh PivotTables and the Data Model, evaluate DAX, run VBA and Python =PY() — and edit your existing workbooks with formulas, PivotTables, charts, macros and formatting left intact. Watch it happen in real time.

🔗 In-Process Service Architecture - The MCP Server hosts the ExcelMcp Service in-process and calls it directly (no pipe), for low-latency Excel automation. The CLI is an equal entry point that runs the same service as a background daemon.

CLI also available: mcp-excel.exe (MCP Server) and excelcli.exe (CLI) are distributed as standalone self-contained executables — no .NET runtime required.

Requirements: Windows OS + Excel 2016+

🚀 Installation

Quick Setup Options:

1. VS Code Extension - One-click install for GitHub Copilot
2. npm - Run through npx with no .NET installation or manual executable setup
3. Standalone exe - Works with Claude Desktop, Cursor, Cline, Windsurf, and other MCP clients
4. MCP Registry - Find us at registry.modelcontextprotocol.io as io.github.sbroenne/mcp-server-excel

Manual Installation (All MCP Clients):

Primary — npm (no .NET runtime required):

npx -y @sbroenne/mcp-server-excel@latest

Configure MCP clients with command: "npx" and args: ["-y", "@sbroenne/mcp-server-excel@latest"]. Requires Node.js 18+; npm resolves @latest at launch using its normal cache policy.

Standalone executable:

# Download ExcelMcp-MCP-Server-{version}-windows.zip from GitHub Releases,
# extract mcp-excel.exe, and configure the client with { "command": "mcp-excel" }.

Secondary — .NET Global Tool (requires .NET 10 runtime):

dotnet tool install --global Sbroenne.ExcelMcp.McpServer

Supported AI Assistants:

• ✅ GitHub Copilot (VS Code, Visual Studio)
• ✅ Claude Desktop
• ✅ Cursor
• ✅ Cline (VS Code Extension)
• ✅ Windsurf
• ✅ Any MCP-compatible client

📖 Detailed setup instructions: MCP Server Installation Guide

🎯 Quick config examples: examples/mcp-configs/

🛠️ What You Can Do

60 MCP tools across 31 feature areas, with 412 operations covering Power Query, Data Model/DAX, What-If Analysis, PivotTables, Excel Tables, Charts, Drawings, VBA, Ranges, Worksheets, Workbooks, QueryTables, XML Maps, Connections, Named Ranges, File/Session management, Calculation Mode, Slicers, Conditional Formatting, Screenshots, and Window Management.

📚 Feature Overview → - Capabilities grouped by goal. Current actions and inputs for all 412 operations are described by the server's tools.

AI-Powered Workflows:

• 💬 Natural language Excel commands through GitHub Copilot, Claude, or ChatGPT
• 🔄 Optimize Power Query M code for performance and readability
• 📊 Build complex DAX measures with AI guidance
• 📋 Automate repetitive data transformations and formatting
• 👀 Show Excel Mode - Say "Show me Excel while you work" to watch changes live

Session workflow

List and match the intended workbook, reuse its session or open/create, perform the work, then list and check its canClose. Close only when authorized and choose save: true or save: false explicitly. Closing without saving discards all unsaved edits and has no tool-level undo.

MCP tools that target an open workbook use workbook_session_id for both inputs and results. This works around a Claude Desktop bridge issue that can drop inputs named session_id; no alias is exposed. CLI JSON keeps its sessionId convention; its sessions are separate.

Calls within a session execute one at a time, but concurrent requests and responses have no guaranteed order. Wait for dependent calls. Writes attempt to restore the calculation mode; restoration can fail without failing the write. Use get-settings when subsequent work depends on the mode. Manual mode requires explicit calculation before relying on dependent values.

💡 Example Use Cases

"Create a sales tracker with Date, Product, Quantity, Unit Price, and Total columns"
→ AI creates the workbook, adds headers, enters sample data, and builds formulas

"Create a PivotTable from this data showing total sales by Product, then add a chart"
→ AI creates PivotTable, configures fields, and adds a linked visualization

"Import products.csv with Power Query, load to Data Model, create a Total Revenue measure"
→ AI imports data, adds to Power Pivot, and creates DAX measures for analysis

"Create a slicer for the Region field so I can filter interactively"
→ AI adds slicers connected to PivotTables or Tables for point-and-click filtering

"Put this data in A1: Name, Age / Alice, 30 / Bob, 25"
→ AI writes data directly to cells using natural delimiters you provide

📋 Additional Resources

• GitHub Repository - Source code, issues, discussions
• MCP Server Installation Guide - Detailed setup for all platforms
• VS Code Extension - One-click installation
• CLI Documentation - the same features as terminal commands for coding agents, usually with fewer AI tokens

License: MIT
Privacy: PRIVACY.md Platform: Windows only (requires Excel 2016+)
Support: GitHub Issues