ironcalc-mcp
Enables AI agents to directly create, read, edit, and save Excel (.xlsx) spreadsheets using natural language through the Model Context Protocol, supporting cells, sheets, formatting, and formulas without writing Python code.
README
ironcalc-mcp
MCP server for the IronCalc spreadsheet engine. Create, read, edit, and save xlsx files through the Model Context Protocol.
Covers the full surface of the IronCalc Python SDK - everything you can do with ironcalc in Python, you can do through MCP.
Motivation
I was really inspired by paper.design's real-time MCP workflow, where exposing a design tool via MCP enables custom, extensible agent workflows that feel native rather than bolted on.
Current AI-spreadsheet integrations (like Claude's cowork mode for Excel) add an unnecessary layer of abstraction - the agent writes Python code that manipulates Excel, instead of talking to the spreadsheet directly. This MCP server removes that indirection. Agents read cells, write formulas, and get evaluated results in one hop. No codegen middleman, no openpyxl glue scripts.
Huge thanks to the IronCalc contributors - their Python SDK API is so clean that this entire MCP server is basically thin wrappers. The hard work is all theirs.
Current limitation: the server works with an in-memory workbook - load a file, edit it, save it back. There's no real-time streaming to a running IronCalc UI yet. The engine already has a diff queue for syncing, so the plumbing is there. If IronCalc adds a WebSocket bridge to their web app, we can wire up live sync and get a true reactive workflow - agent edits appearing in the browser as they happen.
Quick start
uvx ironcalc-mcp
Claude Code
claude mcp add ironcalc -- uvx ironcalc-mcp
Claude Desktop / Cursor
Add to your MCP config:
{
"mcpServers": {
"ironcalc": {
"command": "uvx",
"args": ["ironcalc-mcp"]
}
}
}
Examples
Build a spreadsheet from scratch
Create a budget with formulas and save it
Audit an existing xlsx for errors
Load quarterly_report.xlsx and find all #NAME?, #REF!, #VALUE! errors. Tell me which formulas are broken and why.
Fix broken cross-sheet references
The P&L sheet was renamed but formulas in the DCF sheet still reference the old name. Find and fix all broken references.
Build a financial model
Create a 5-year DCF model with revenue assumptions in one sheet, P&L in another, and free cash flow calculation in a third. Use =NPV() and =IRR() for valuation.
Analyze and transform data
Load sales_data.xlsx, add a column with running totals, insert a summary row at the bottom, and save.
Available tools (21)
Workbook
| Tool | Description |
|---|---|
create_workbook |
Create a new empty workbook. Params: name, locale, timezone, language |
load_workbook |
Load workbook from an xlsx file. Params: file_path, locale, timezone, language |
save_workbook |
Save workbook to xlsx. Params: file_path (optional, defaults to load path) |
get_workbook_info |
Get sheets list, dimensions, state, colors |
Cells
| Tool | Description |
|---|---|
get_cell |
Get content, formatted value, and type of a single cell |
get_cells |
Read a rectangular range of cells |
set_cell |
Set a cell's value or formula (prefix with =) |
set_cells |
Batch set multiple cells in one call |
Sheets
| Tool | Description |
|---|---|
add_sheet |
Create a new sheet (auto-named or custom name) |
delete_sheet |
Remove a sheet by index |
rename_sheet |
Rename a sheet |
set_sheet_color |
Set the tab color (hex) |
Structure
| Tool | Description |
|---|---|
insert_rows |
Insert rows at a position |
delete_rows |
Delete rows at a position |
insert_columns |
Insert columns at a position |
delete_columns |
Delete columns at a position |
set_column_width |
Set column width in points |
set_row_height |
Set row height in points |
set_frozen_rows |
Freeze top N rows |
set_frozen_columns |
Freeze left N columns |
Formatting
| Tool | Description |
|---|---|
get_cell_style |
Get font, fill, border, alignment, number format |
set_cell_styleis not yet available — the IronCalc Python binding exposes style properties as read-only. PRs welcome to add#[pyo3(get, set)]upstream.
Contributing
Contributions are welcome! To get started:
git clone https://github.com/yolonir/ironcalc-mcp
cd ironcalc-mcp
uv sync
uv run ironcalc-mcp
Please create a new branch for your changes and open a pull request against main. Pls keep them focused.
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。