Skip to content

Summary

Create, read and edit Excel workbooks from an agent — charts, real PivotTables, slicers, formulas and conditional formatting — with no Microsoft Excel installed, on Windows, macOS or Linux.

Features

  • Charts, refreshable PivotTables, slicers, timelines and sparklines
  • Built-in formula engine with 479 functions and dynamic arrays — no Excel required
  • Trace precedents and dependents, step-through formula evaluation and Goal Seek
  • Streaming, paged reads for sheets with hundreds of thousands of rows
  • --tools, --read-only and --allow-dir to limit context cost and file access

Installation

Set up this MCP server in your favorite AI agent — copy a ready-made configuration below.

Any MCP-compatible agent

Most agents (Claude, Cursor, Windsurf, VS Code, and more) read a standard mcpServers configuration.

  1. Open your agent's MCP configuration file.
  2. Merge the snippet below into it.
  3. Restart the agent — the "Excel MCP Server" tools will be available.
{
  "mcpServers": {
    "excel-mcp-server": {
      "command": "uvx",
      "args": [
        "excel-mcp-server",
        "stdio",
        "--allow-dir",
        "/path/to/workbooks"
      ]
    }
  }
}

Description

Excel MCP Server works on .xlsx, .xlsm, .xltx and .xltm files directly, built on openpyxl, so an agent can produce a finished workbook on a machine — or a server — that has never had Excel installed. It goes well beyond writing cells: charts (including the Excel 2016 types such as waterfall, histogram, Pareto, treemap and sunburst), PivotTables that Excel refreshes, slicers and timelines, sparklines, tables with totals rows, conditional formatting and data validation.

Formulas without Excel. A built-in calculation engine evaluates 479 functions — checked, per the README, against more than 25,000 formulas recorded from real Excel — so read_range returns computed values for files that were never opened in Excel. Dynamic arrays (FILTER, SORT, UNIQUE, LET) are stored the way Excel stores them, and the server can trace precedents and dependents, step through a formula like Excel's Evaluate Formula, and run Goal Seek.

Sized for real files and real context windows. Reads stream the workbook and page their results, and edits to large sheets parse only the rows they touch. Every tool costs context, so --tools loads only the core read tools plus the areas you name (data, charts, pivots, formatting and so on), and the README publishes the character cost of each area. VBA in .xlsm files can be read; writing it is opt-in with --allow-vba-write, and the server never runs macros.

Safety options include folder confinement with --allow-dir, --read-only, a formula safety check, atomic saves and size limits. It runs over stdio or Streamable HTTP (localhost-only unless a bearer token is set), with a Claude Desktop .mcpb bundle and Docker image available. MIT-licensed, by Haris Musa.

Featured

Official AntV MCP server that turns a description of your data into a rendered chart — 26 visualization types from bar and line to sankey, mind-map, fishbone and geographic maps.

MCP: MCP Atlassian

by sooperset

New

Community MCP server for Jira and Confluence with 98 tools, supporting both Atlassian Cloud and self-managed Server/Data Center.

MCP: LinkedIn MCP Server

by stickerdaniel

New

Open-source MCP server that works LinkedIn through your own browser session: profiles, companies, job search, posts, messages and connection requests.

MCP: DaVinci Resolve MCP

by Samuel Gursky

New

Community MCP server that lets AI assistants control DaVinci Resolve through its official scripting API: editing, media pool, grading, Fusion, Fairlight, render setup and review markers.

Browse all MCP servers →