🤖 AI Summary
            spreadsheet-read-mcp is a new high-performance, Rust-based Model Context Protocol (MCP) server that provides deterministic, read-only inspection of spreadsheet workbooks for LLM agents. It’s designed to let models safely explore structure, formulas, styles, named ranges and statistics without mutating files or executing macros — prioritizing high-signal slices (pages, formula summaries, cached values) over full-file loads. The project targets XLSX first (parsed with umya-spreadsheet), can discover XLS/XLSB files, and exposes stable short IDs and manifest stubs so downstream agent harnesses can reference and ingest workbook metadata reliably.
Technically, the server streams MCP over Server-Sent Events by default (SSE) and supports HTTP or stdio transports, configurable by CLI, env or YAML/JSON. Core capabilities include paged sheet streaming, formula clustering, precedent/dependent tracing with pagination, volatile-function scanning, sheet classification, style reuse summaries, named ranges, and search-by-regex. A bounded LRU workbook cache (configurable) keeps hot files in memory; unsupported structures are reported as MCP errors rather than crashing. Important limitations: strictly read-only (no writing, recalculation, VBA/macro execution or on-the-fly format conversion), and backends beyond XLSX require future adapters. The repo includes tests, CI, docs on architecture/roadmap, and a cargo-installable binary for local use.
        
            Loading comments...
        
        
        
        
        
            login to comment
        
        
        
        
        
        
        
        loading comments...
        no comments yet