LogiSheets

What LogiSheets MCP server does

The LogiSheets MCP server connects an MCP host to the LogiSheets spreadsheet engine, which is written in Rust and runs on the local machine without opening network sockets. It can start a blank workbook or load an existing .xlsx, calculate formulas, modify workbook contents, and save a genuine .xlsx file for another person to open and continue using.

Its main abstraction is the block: a named table with fields, a key field for rows, optional descriptions, and rules for calculated or validated fields. This lets an agent address data by block, row key, and field instead of relying only on changing cell coordinates. Existing tables can be adopted with convert_to_block, while new tables can be created with create_block.

How it works

A typical workflow opens or creates a workbook, inspects its blocks, defines field rules, adds or edits rows, evaluates formulas, and saves the result. describe_block returns a block’s schema and can include values; list_blocks provides a workbook-level view. Formula evaluation is non-destructive: eval_formula returns a result without storing it.

Block edits include adding, deleting, moving, and atomically updating rows or cells. Field rules can define formulas, validation behavior, and editability. Analysis blocks and pivots derive additional views from source data, while refresh and edit operations keep those structures aligned with their sources. trace follows cell dependencies in either direction, and list_violations identifies cells that fail validation rules.

The LogiSheets MCP server also supports hypothetical edits. preview_changes can test one proposed change or a grid of scenarios without applying them, and goal_seek searches for an input that produces a target output. Defined names can be inspected with list_names, and sheets can be added with create_sheet.

Setup and configuration

The package requires Node 20 or newer. Install it globally with:

npm install -g logisheets-mcp

A workbook is optional at startup: the first use can create one, or open_workbook can load an existing .xlsx.

Use save_workbook when the resulting workbook should be written to a shared filesystem. For hosts that cannot access the same filesystem, export_xlsx returns the file as base64 instead.

Tools and capabilities

The LogiSheets MCP server exposes tools for:

• Workbook and sheet lifecycle: open, save, export, and create sheets.
• Structured tables: create or adopt blocks, describe them, document their purpose, and list all blocks.
• Data editing: add, delete, reorder, and batch-update rows and cells.
• Calculation: evaluate formulas, define field rules, inspect dependencies, preview scenarios, and perform goal seeking.
• Derived views: create or edit analysis blocks and pivots, then refresh pivots from their source blocks.
• Validation and metadata: list rule violations and defined names.
• Charts: create charts from block fields or arbitrary A1 ranges, list chart definitions, update charts, and delete them.

Charts are stored in the workbook and refer to source ranges or block fields rather than storing only a rendered image. chart_from_block follows fields as block rows change; chart_insert handles data that remains in ordinary cell ranges.

Limitations and notes

The server is local and uses stdio rather than a network socket, so deployment details differ from an HTTP MCP service.

Block schemas describe structure and field rules; descriptions provide the prose context for what rows and fields mean. That context is stored in the workbook and can be recovered in a later session. Reading schema information does not replace reading the underlying data when the agent needs the actual records.

eval_formula does not persist its result. Row movement is presentation-only and does not change values. For pivots, edit_pivot is the supported way to repair a broken pivot, while refresh_pivot brings its rows and columns back into line with the source. The project is licensed under MIT.