Most spreadsheet servers stop at reading cells; this one carries through to pivot tables, DCF and variance models, and multi-sheet output with formulas intact. Two details decide whether it fits your files: the chunked reader exists for workbooks too big to return in one go, and a workspace root variable pins every read and write to a single directory when you want that boundary.
A local file server for .csv, .xlsx and .xls: reading and cell addressing at one end, statistics, pivots, financial models and multi-sheet writing at the other.
- Reading at every grain: a whole file, one A1-notation cell, a range, the header row, or a keyword search across cells
- Filtering and aggregation on a column, plus pivot tables with grouping
- Statistics per column, correlation between two numeric columns, and a profiling report across all columns
- Financial models on your own data: DCF valuation, budget versus actual variance, ratio analysis, what-if scenarios and trend analysis with growth rates and seasonality
- Writing back out: new files, extra sheets, multi-sheet workbooks with formulas and inter-sheet references, cell formatting and column auto-fit
- Cleanup passes — duplicate detection, batch data cleaning, and a VLOOKUP helper with fuzzy matching across two files
- Work spanning several files at once: aggregate or filter across a list of them, and cross-validate referential integrity between them
- Chunked reads with a sizing check first, for files too large to return in one response
No account. Installed from npm as excel-csv-mcp-server and run over stdio as a local process with your own user's permissions. The AI-backed formula tools want one provider key — ANTHROPIC_API_KEY, OPENAI_API_KEY, DEEPSEEK_API_KEY or GEMINI_API_KEY — and fall back to a local provider that makes no network calls when none is set. Setting EXCEL_MCP_WORKSPACE_ROOT confines every read and write to one directory.
One command — npm install -g excel-csv-mcp-server