Mcp Server Spreadsheet

mcp-server-spreadsheet

mcp-name: io.github.marekrost/mcp-server-spreadsheet

Data-first MCP server for reading and writing spreadsheet files (.xlsx, .csv, .ods).

Key features

• Multi-format — works with Excel (.xlsx), CSV (.csv), and OpenDocument (.ods) files through a unified tool interface.
• Dual mode — cell-level workbook operations and a DuckDB-powered SQL query engine, interleaved freely on the same file.
• Workbook essentials — worksheets, rows, columns, cells, search.
• Data-only — preserves existing formatting but only reads and writes values.
• Stateless — every call specifies file and sheet explicitly; no handles or sessions.
• Atomic saves — writes go to a temp file, then os.replace() into the target path.
• Type coercion on write — numeric strings become numbers, everything else is text.
• SQL across sheets — JOINs, GROUP BY, aggregates, subqueries via in-memory DuckDB; mutations write back to the file.
• CSV as single-sheet workbook — CSV files are treated as a workbook with one sheet named default.

Requirements

• Python 3.10+

Installation

From PyPI (recommended)

No local checkout needed — just configure your MCP client (see below).

From source (for development)

git clone https://github.com/marekrost/mcp-server-spreadsheet.gitcd mcp-server-spreadsheetuv sync

Usage

Claude Desktop

Add to your claude_desktop_config.json:

Using PyPI (recommended):

{  "mcpServers": {    "mcp-server-spreadsheet": {      "command": "uvx",      "args": ["mcp-server-spreadsheet"]    }  }}

Using local source:

{  "mcpServers": {    "mcp-server-spreadsheet": {      "command": "uv",      "args": ["run", "--directory", "/path/to/mcp-server-spreadsheet", "main.py"]    }  }}

Claude Code

Add to your .mcp.json:

Using PyPI (recommended):

{  "mcpServers": {    "mcp-server-spreadsheet": {      "command": "uvx",      "args": ["mcp-server-spreadsheet"]    }  }}

Using local source:

{  "mcpServers": {    "mcp-server-spreadsheet": {      "command": "uv",      "args": ["run", "--directory", "/path/to/mcp-server-spreadsheet", "main.py"]    }  }}

Standalone (stdio transport)

# PyPIuvx mcp-server-spreadsheet
# Local sourceuv run main.py

Restricting file access to a directory (optional)

Set MCP_SPREADSHEET_ROOT to confine all path arguments to a single directory tree. Paths outside it are rejected with a clear error returned to the agent.

{  "mcpServers": {    "mcp-server-spreadsheet": {      "command": "uvx",      "args": ["mcp-server-spreadsheet"],      "env": { "MCP_SPREADSHEET_ROOT": "/home/me/spreadsheets" }    }  }}

Unset (the default), any path the server process can access is allowed.

Format notes

Sheet management tools (add_sheet, delete_sheet, copy_sheet) raise an error for CSV files.

Tools

Workbook Operations

Sheet Operations

Reading Data

Writing Data

Column Operations

Search

Table Mode (SQL)

SQL examples:

-- Filter and sortSELECT name, revenue FROM Sales WHERE status = 'Active' ORDER BY revenue DESC LIMIT 20
-- Cross-sheet JOINSELECT o.order_id, c.name FROM Orders o JOIN Customers c ON o.customer_id = c.id
-- AggregateSELECT department, COUNT(*) AS n, AVG(salary) AS avg FROM Employees GROUP BY department
-- MutateUPDATE Sales SET status = 'Closed' WHERE quarter = 'Q1' AND revenue < 1000DELETE FROM Logs WHERE date < '2024-01-01'

Sheet names with spaces must be quoted: SELECT * FROM "Q1 Sales".

Sheets whose table doesn't start at row 1

All three SQL tools accept header_row and data_start_row. Each can be an int (applied to every sheet) or a {sheet_name: row} mapping (sheets not listed fall back to the default). Use header_row when column titles live below row 1, and data_start_row when extra rows (e.g. a units row) sit between the header and the data.

# Header on row 3, data follows immediatelysql_query(file, 'SELECT * FROM "People"', header_row=3)
# Mixed workbook: People headers at row 3, Orders header at row 1 with a# units row at row 2.sql_query(    file,    'SELECT * FROM "Orders" o JOIN "People" p ON o.name = p.name',    header_row={"People": 3, "Orders": 1},    data_start_row={"Orders": 3},)

sql_execute preserves rows above header_row when writing changes back.

Running tests

uv sync --group devuv run pytest

Every tool is exercised against .xlsx, .csv, and .ods fixtures generated into a temp directory.

Common Parameters

Every sheet-level tool accepts:

All row/column indices are 1-based. Cell references use A1 notation (A1, $B$2).