What problem does it solve? Running ArcPy geoprocessing from an AI assistant normally means giving the agent unrestricted access to a licensed ArcGIS Pro runtime and the local filesystem. This Skill bridges MCP clients like Claude Desktop to ArcGIS Pro through a controlled local server that validates every input, sandboxes all filesystem paths, and isolates ArcPy in a separate worker process so native crashes never take down the MCP host. ## Core Features & Use Cases - 100-tool ArcPy catalog: Covers 10 GIS verticals including data management, geometry analysis (buffer, clip, intersect, dissolve), raster operations (slope, zonal statistics, hydrology), network analysis, spatial statistics, and layout export to PDF/PNG. - PathGuard sandbox: Every filesystem argument is resolved and containment-checked against configured allowed roots in both the server and worker processes, rejecting traversal, UNC, and out-of-root paths. - Destructive operation gates: Ten state-mutating tools (delete, append, repair geometry, calculate field, and others) require an explicit confirm=true token before execution. - Use Case: Ask Claude to buffer all parcels in a geodatabase by 50 meters, project the result to a target EPSG code, add it to an .aprx map, and export the layout as a PDF, all within a confined workspace directory. ## Quick Start Install the package with pip, run arcgis-mcp-setup to clone the ArcGIS Pro Python environment, point ARCPY_PYTHON_PATH at the reported interpreter, and ask your MCP client to run health_check before any geoprocessing.