QGIS

Translates natural language prompts from Claude into executable QGIS operations by implementing the Model Context Protocol (MCP) as a bridge layer. Claude interprets user intent and maps it to specific tool calls (create_new_project, add_vector_layer, etc.) which are then relayed through the MCP server to the QGIS plugin for execution. This enables users to describe geospatial tasks in plain English rather than writing PyQGIS code directly.

Implements a persistent socket server within the QGIS plugin that receives JSON-serialized commands from the MCP server and executes them using PyQGIS APIs. The plugin maintains a listening socket on localhost, parses incoming command payloads, executes the corresponding PyQGIS operation, and returns structured JSON responses. This architecture decouples Claude's reasoning from QGIS execution, allowing asynchronous command processing without blocking the QGIS UI.

Provides Claude with tools to manage QGIS project files through create_new_project, load_project, save_project, and get_project_info commands. These operations directly invoke PyQGIS QgsProject APIs to manipulate the project state, including creating blank projects, loading .qgs/.qgz files from disk, persisting changes, and retrieving metadata like CRS, extent, and layer count. All operations return structured metadata enabling Claude to reason about project state.

Enables Claude to manipulate layers in the active QGIS project through add_vector_layer, add_raster_layer, remove_layer, get_layers, zoom_to_layer, and get_layer_features commands. These tools invoke PyQGIS layer APIs to load data sources (shapefiles, GeoTIFFs, PostGIS, etc.), manage the layer tree, retrieve feature data with optional filtering, and adjust the map canvas extent. Layer operations return structured metadata (layer IDs, geometry types, feature counts) enabling Claude to chain operations.

Provides an execute_code tool that allows Claude to run arbitrary PyQGIS Python code strings directly within the QGIS environment. The code is executed in the context of the QGIS plugin with access to the current project, layers, and canvas. Execution results and errors are captured and returned as structured responses, enabling Claude to perform custom spatial operations not covered by the standard tool set. This is a powerful escape hatch for advanced workflows.

Exposes QGIS's processing framework through an execute_processing tool that allows Claude to invoke any registered processing algorithm (from QGIS core, GDAL, SAGA, etc.) with structured parameter binding. Claude specifies the algorithm ID and parameters as a dictionary, which are validated and passed to the processing engine. Results include output layer paths, statistics, and execution status. This enables Claude to leverage QGIS's extensive algorithm library without custom code.

Provides a render_map tool that captures the current QGIS map canvas as a raster image file (PNG, JPEG, etc.) with the current symbology, labels, and extent. The rendering is performed by QGIS's rendering engine, ensuring visual fidelity. Claude can use this to generate visualizations for analysis results, create map exports for reports, or verify that layer operations produced expected visual results. Supports custom output paths and image formats.

Provides ping and get_qgis_info tools for monitoring the health and status of the QGIS MCP integration. The ping command performs a simple round-trip test to verify socket connectivity between the MCP server and QGIS plugin. The get_qgis_info command returns metadata about the QGIS installation (version, plugins, available providers, etc.), enabling Claude to adapt its behavior based on available capabilities. These tools are essential for debugging and ensuring reliable operation.

Implements the Model Context Protocol (MCP) specification in the QgisMCPServer class, which exposes all QGIS operations as standardized MCP tools with JSON schemas. The server maintains a tool registry that Claude can query to discover available operations, their parameters, and return types. This enables Claude to understand the QGIS API surface without hardcoding tool knowledge. The MCP implementation handles protocol marshaling, error serialization, and response formatting according to the MCP specification.