Skip to main content
Glama
sahan411

Spreadshet Audit MCP

by sahan411
OverviewSchemaSourceRelated ServersDiscussionsAdmin
Python
Local

spreadsheet-audit-mcp

An MCP server that lets Claude (or any MCP client) audit an .xlsx workbook for the errors that quietly break spreadsheet models. Local and read-only: it opens the file you point it at, never writes to it, never uses the network.

What it checks

Rule

Severity

What it catches

CALC-ERROR

HIGH

#DIV/0!, #REF!, #N/A, #VALUE!, #NAME? after an independent recalculation (pure-Python formulas engine)

CACHED-ERROR

HIGH

error values stored in the file by the last app that saved it

TYPED-OVER

HIGH

a hard-coded number sitting inside a filled-down formula run

PATTERN-BREAK

HIGH

the one formula that differs from consistent neighbours above and below

EXT-LINK

HIGH

formulas that read another file

NAME-REF

HIGH

named ranges that point to #REF!

MAGIC-NUMBER

MEDIUM

rates and thresholds buried in formulas instead of labelled inputs (rounding tolerances like <=0.005 are ignored)

LOOKUP-APPROX

MEDIUM

VLOOKUP with approximate match (omitted or TRUE)

HIDDEN-SHEET

MEDIUM

hidden or very-hidden sheets

HIDDEN-ROWCOL, VOLATILE

LOW

hidden rows/columns; INDIRECT, OFFSET, TODAY, NOW, RAND

Repeated MAGIC-NUMBER, VOLATILE and LOOKUP-APPROX hits are grouped per sheet with a count, so 700 repeats do not bury the real problems.

Related MCP server: sheetspec

Tools

Tool

What it does

audit_workbook

Full audit (path, recalc, min_severity, limit).

inspect_cell

One cell's formula, stored result and the cells it reads (path, sheet, cell).

Install and use

Requires Node 18+, Python 3.9+ with openpyxl (and optionally formulas for the recalculation check): pip install openpyxl formulas.

npm install
npm test        # 14 Python tests + 8 end-to-end MCP client checks

Claude Desktop / Claude Code config:

{ "mcpServers": { "spreadsheet-audit": { "command": "node", "args": ["/absolute/path/to/server.mjs"] } } }

Then ask: "Audit ./budget.xlsx and tell me which findings could change the bottom line."

Honest limits

Heuristic. It points at where models usually go wrong; it cannot prove a model is right, and some findings are intentional (for example a sensitivity grid with fixed steps). If the recalculation engine cannot handle a workbook it says so in recalculation_note and the static checks still run.

Need a spreadsheet model built or reviewed, or a custom MCP server for your tools?

Open an issue or message the author.

License: MIT

This server cannot be deployed

Shareable, User-Owned Context BackendPeople, applications, and agents read, write, and coordinate together in one context lake. Reach it over MCP, a CLI, a Python SDK, or REST.fulcradynamics.comAd

Maintenance

ActivityMaintained
ResponsivenessNo issues

Resources

Related MCP Connectors

Related MCP Servers

  • xlsx-for-ai

    senoff
    A
    license
    A
    quality
    A
    maintenance
    50 tools and 400 functions for working with Excel/.xlsx spreadsheets — read/write, recalculate formulas, diff, repair broken references, and audit. Built for AI agents.
    50
    949 npm
    6
    MIT

  • sheetspec

    helloo1568
    A
    license
    Not graded
    quality
    A
    maintenance
    Validates Excel (.xlsx) workbooks against reviewed, lockable acceptance contracts, catching errors and outputting structured issues that agents can repair.
    18 PyPI
    1
    MIT

  • excel-mcp

    Matsu132
    F
    license
    C
    quality
    B
    maintenance
    An MCP server for safe .xlsx inspection and controlled workbook writes, enabling spreadsheet reads, search, filtering, aggregation, and preview-based writes with workspace confinement and audit logging.
    12
    -