GDAL MCP is a Model Context Protocol (MCP) server for geospatial AI workflows. It enables AI agents to perform raster and vector analysis through a comprehensive toolset while enforcing justification of methodological decisions via a reflection middleware system. The reflection system captures structured reasoning before execution, including intent, alternatives, tradeoffs, and confidence, and caches it to support stateful, cross-domain workflows. This cross-domain cache sharing lets reasoning and justifications developed for raster operations influence subsequent vector tasks, preserving methodological continuity between data types. Built on a Python-native stack (Rasterio, PyProj, pyogrio, Shapely, NumPy) and exposed through the FastMCP Context API, the server provides real-time feedback and audit trails. It supports 12 production-ready tools across raster and vector domains (info, convert, reproject, stats; info, reproject, convert, clip, buffer, simplify) with reflection-enabled first-use prompts and instant re-execution on cache hits. Additional MCP resources include a workspace catalog, metadata intelligence, and a reference knowledge base to promote reproducible geospatial science in production-grade environments, with workspace security via path validation.