Geography shapes nearly every decision humans make — where to build, what to grow, how to respond to disasters, which routes to drive. Yet for all the power of modern GIS tools, they remain stubbornly specialized. You need years of training to operate ArcGIS Pro effectively. Writing a PostGIS query that correctly handles coordinate reference systems requires deep expertise. Analyzing satellite imagery demands knowledge of spectral bands, atmospheric correction, and classification algorithms.

The Model Context Protocol is changing this by making geospatial capabilities accessible to AI agents through standardized tool interfaces. Instead of learning GIS software, you describe what you need in natural language, and an MCP-connected agent handles the spatial operations, map rendering, satellite data retrieval, and coordinate transformations behind the scenes. The geospatial AI market is projected to reach $45 billion in 2026 (Precedence Research), and MCP is becoming the standard interface through which AI agents access spatial intelligence.

This guide covers the geospatial MCP ecosystem — GIS operation servers, desktop GIS bridges, cloud mapping platforms, satellite imagery and earth observation, location services, and the security considerations specific to spatial data. Our analysis draws on published documentation, open-source repositories, vendor announcements, and industry reports — we research and analyze rather than deploying these systems ourselves. Rob Nugen operates ChatForest; the site’s content is researched and written by AI.

Why Geospatial Needs MCP

Geospatial work has properties that make it both a natural fit for AI agents and uniquely challenging:

Specialized knowledge barriers. GIS operations involve coordinate reference systems (there are thousands), spatial algorithms (buffering, intersection, union, Voronoi diagrams), data formats (GeoJSON, Shapefile, GeoPackage, GeoTIFF, GeoParquet), and domain-specific tools. Most developers can’t write a correct spatial join without significant learning.

Data scale and diversity. A single satellite image can be gigabytes. Global mapping datasets like Overture Maps contain billions of features. Combining street-level data with elevation models, land use classifications, and real-time traffic requires merging radically different data sources.

Real-time requirements. Navigation, fleet management, disaster response, and precision agriculture all need current data — not yesterday’s snapshot. AI agents need to query live mapping services, not just static files.

Multi-tool workflows. A typical geospatial analysis chains together geocoding, spatial queries, raster analysis, network routing, and visualization. No single tool does everything, making MCP’s tool composition model particularly valuable.

MCP provides a standard protocol for AI agents to discover and use geospatial tools across this fragmented landscape — from open-source Python libraries to enterprise cloud platforms to desktop GIS applications.

GIS Operations Servers

These MCP servers provide direct access to core geospatial operations — the geometry calculations, coordinate transformations, and spatial analysis functions that form the foundation of GIS work.

GIS-MCP (mahdin75)

Repository: mahdin75/gis-mcp (~120 stars) | License: MIT | Language: Python

GIS-MCP is the most comprehensive open-source geospatial MCP server, providing 89 functions across multiple categories by wrapping established Python GIS libraries:

  • Shapely — geometry operations (intersection, union, buffer, difference, convex hull, Voronoi diagrams, simplification)
  • PyProj — coordinate reference system transformations between any of the thousands of CRS definitions
  • GeoPandas — spatial data manipulation, joins, overlays, and file format conversion
  • Rasterio — raster data processing (reading GeoTIFFs, extracting values, band math)
  • PySAL — spatial statistics and econometrics (spatial autocorrelation, clustering)

The 89 tools span several categories:

Category Examples
Geometry operations Intersection, union, buffer, difference, symmetric difference, convex hull
Coordinate transforms Reproject between CRS, transform geometry coordinates
Measurements Distance, area, length, centroid, bounds calculations
Spatial analysis Proximity checks, spatial overlays, spatial joins, validation
Data conversion GeoJSON, WKT, Shapefile, GeoPackage format transformations
Raster processing Band extraction, value sampling, zonal statistics

The server supports both STDIO transport (for local use with Claude Desktop or Cursor) and HTTP/SSE transport with RESTful storage endpoints for file upload/download. This dual transport model makes it suitable for both desktop AI workflows and server-side applications.

GeoServer MCP (mahdin75)

Repository: mahdin75/geoserver-mcp | License: MIT | Language: Python

From the same developer, GeoServer MCP bridges AI agents to GeoServer — the leading open-source server for sharing geospatial data via OGC standards (WMS, WFS, WCS). This is particularly valuable for organizations that already serve spatial data through GeoServer:

  • Workspace management — list, create, and configure GeoServer workspaces
  • Layer operations — query available layers, their schemas, and spatial extents
  • Feature querying — execute spatial queries against vector data with CQL filters
  • Map generation — request rendered maps via WMS with style control
  • WFS access — retrieve features in GeoJSON format for AI processing

The Docker installation provides the quickest setup. Version 0.5.0 (beta) is under active development.

Desktop GIS Bridges

These servers connect AI agents to established desktop GIS applications, enabling natural language control of professional GIS workflows.

QGIS MCP (jjsantos01)

Repository: jjsantos01/qgis_mcp (~623 stars, 84 forks) | License: Open source | Language: Python

QGIS MCP is the most popular geospatial MCP server by star count, connecting Claude (or any MCP client) to QGIS Desktop — the leading open-source GIS application. The architecture uses a socket-based plugin system:

  1. QGIS Plugin — runs inside QGIS as a server, receiving and executing commands
  2. MCP Server — implements the Model Context Protocol and bridges to the QGIS plugin via socket

Available tools include:

  • Project management — create new projects, load existing ones, get project info
  • Layer operations — add vector layers, add raster layers, list layers, remove layers, zoom to layer extent
  • Processing algorithms — execute any algorithm from QGIS’s Processing Toolbox (hundreds of spatial analysis tools)
  • Code execution — run arbitrary Python code within QGIS’s environment
  • Info retrieval — query QGIS version, installed plugins, and system capabilities

The Processing Toolbox access is particularly powerful — it exposes QGIS’s entire library of spatial algorithms (GDAL, GRASS, SAGA) through a single MCP tool, giving AI agents access to hundreds of geospatial operations.

ArcGIS Pro MCP Server (nicogis)

Repository: nicogis/MCP-Server-ArcGIS-Pro-AddIn | Language: C# (.NET 8)

For organizations using Esri’s commercial ArcGIS Pro, this community-built server bridges MCP to ArcGIS Pro through a .NET add-in:

  • Architecture: ArcGIS Pro Add-In (C# with ArcGIS Pro SDK) communicates with an MCP Server (.NET 8 console app) via Named Pipes (IPC)
  • Capabilities: List layers, count features, zoom to layers, query map state, execute geoprocessing tools
  • Security: All communication is local via Named Pipes — no network exposure
  • Integration: Configured through Visual Studio’s .mcp.json for use with VS Code Copilot Agent Mode

This is a community project, not officially supported by Esri, but it demonstrates how desktop GIS applications can become MCP-accessible for AI-assisted spatial analysis.

Cloud Mapping Platforms

Enterprise mapping platforms have embraced MCP as the interface for their agentic GIS strategies.

CARTO MCP Server

Provider: CARTO (commercial, self-described “Agentic GIS Platform”) | Type: Remote MCP server

CARTO’s approach is distinctive — rather than exposing individual API endpoints as MCP tools, they expose entire Workflows (visual processing pipelines) as MCP tools. Each workflow defines a complete spatial analysis operation:

  • Spatial operations — buffers, spatial joins, geocoding, routing
  • Advanced analytics — site selection analysis, routing optimization, catchment area calculation
  • Machine learning — predictive analytics powered by Geospatial Foundation Models
  • Earth Engine integration — Google Earth Engine datasets accessible through CARTO workflows
  • Custom extensions — organizations build and expose their own spatial analysis pipelines

The workflow-as-tool approach provides several advantages:

┌──────────────┐     MCP      ┌──────────────────────┐
│   AI Agent   │◄────────────►│   CARTO MCP Server   │
│  (Claude,    │              │                      │
│   Gemini,    │              │  ┌────────────────┐  │
│   ChatGPT)   │              │  │   Workflow 1:   │  │
│              │              │  │  Site Selection │  │
│              │              │  ├────────────────┤  │
│              │              │  │   Workflow 2:   │  │
│              │              │  │  Route Optimize │  │
│              │              │  ├────────────────┤  │
│              │              │  │   Workflow 3:   │  │
│              │              │  │  Catchment Area │  │
│              │              │  └────────────────┘  │
│              │              │          │           │
│              │              │          ▼           │
│              │              │  ┌────────────────┐  │
│              │              │  │  Data Warehouse │  │
│              │              │  │  (BigQuery,     │  │
│              │              │  │   Snowflake,    │  │
│              │              │  │   Databricks,   │  │
│              │              │  │   PostgreSQL)   │  │
│              │              │  └────────────────┘  │
└──────────────┘              └──────────────────────┘

Agents get transparency (each workflow is a visible, version-controlled pipeline), governance (CARTO controls what data and operations are exposed), and composability (agents can chain workflows for multi-step spatial analysis). CARTO recently announced integration with Oracle Generative AI, expanding the range of AI platforms that can use their MCP server.

Mapbox MCP Server

Repository: mapbox/mcp-server | Provider: Mapbox (commercial)

Mapbox’s official MCP server provides AI agents with structured access to the full Mapbox location platform:

  • Geocoding — convert addresses and place names to coordinates (and reverse)
  • Navigation — driving, walking, and cycling directions with real-time and typical traffic
  • Matrix API — calculate travel times between multiple origins and destinations for logistics optimization
  • Isochrone API — generate areas reachable within specific time or distance constraints
  • Map rendering — static map tiles in multiple styles

Mapbox’s 2026 predictions identify three families of specialized geospatial agents in development: Feedback Agents (processing user map corrections), Navigation Agents (in-vehicle route optimization), and Location Agents (general spatial reasoning). They predict that by 2026, geospatial tools will shift from “add-ons” to built-in capabilities in major LLMs — similar to how Google is integrating Maps directly into Gemini.

ArcGIS Location Services MCP

Repository: puran1218/arcgis-location-services-mcp | Requires: ArcGIS Developer API key

This MCP server wraps Esri’s ArcGIS Location Platform — a managed PaaS providing curated global geospatial data:

  • Geocoding — forward and reverse geocoding with global coverage (2026 updates improve house-number accuracy in 8+ additional regions)
  • Places — find points of interest near specified locations
  • Routing — turn-by-turn directions between locations
  • Elevation — retrieve elevation data for specified coordinates
  • Basemap tiles — access static basemap tiles in various styles

Unlike the ArcGIS Pro bridge (which connects to desktop software), this server connects to Esri’s cloud services — no local GIS installation required, but it needs an API key and consumes platform credits.

Earth Observation and Satellite Imagery

MCP servers for earth observation connect AI agents to satellite data from NASA, ESA, and Google Earth Engine — enabling natural language access to petabytes of imagery.

Axion Planetary MCP

Repository: Dhenenjay/Axion-MCP | Language: Node.js

Axion Planetary MCP connects AI agents to Google Earth Engine — Google’s platform for petabyte-scale satellite data analysis. This is arguably the most powerful geospatial MCP server by capability:

  • 100+ satellite datasets — Landsat, Sentinel, MODIS, VIIRS, and more
  • 30+ analysis tools — vegetation indices (NDVI, NDWI), terrain analysis, spectral calculations
  • 5 pre-trained ML models — wildfire risk assessment, flood prediction, agriculture health monitoring, deforestation detection, water quality analysis
  • Interactive maps — generate and render analysis results as viewable maps
  • Temporal analysis — track changes across time using multi-date imagery composites

Use cases include crop health monitoring for precision agriculture, deforestation tracking for environmental organizations, flood extent mapping for disaster response, and urban expansion analysis for city planning. The platform is designed to make Earth observation accessible to non-specialists — “from farmers to NGOs” — through natural language queries.

Note: Requires Google Earth Engine credentials (free for research and education, paid for commercial use).

Copernicus Earth Observation MCP

Repository: wb1016/copernicus-mcp | Language: Python

This server provides access to the European Space Agency’s Copernicus Data Space ecosystem — the world’s largest provider of free satellite imagery:

  • Multi-mission support — Sentinel-1 (radar), Sentinel-2 (optical), Sentinel-3 (ocean/land), Sentinel-5P (atmosphere), Sentinel-6 (sea level)
  • Advanced search — filter by location, date range, cloud cover percentage, and mission-specific parameters
  • Download options — full products, quicklook previews, and batch operations
  • File management — automated cleanup strategies based on age or size

The Copernicus programme provides global coverage with frequent revisits (Sentinel-2 covers the entire Earth every 5 days), making it essential for monitoring applications. The MCP server’s search and download tools let AI agents programmatically find and retrieve imagery matching specific criteria.

NASA MCP Servers

Multiple implementations connect AI agents to NASA’s public data APIs:

ProgramComputer/NASA-MCP-server provides the broadest coverage with 20+ data sources:

  • APOD — Astronomy Picture of the Day
  • Mars Rover Photos — imagery from Curiosity, Perseverance, Opportunity, and Spirit
  • EPIC — Earth Polychromatic Imaging Camera (full-disk Earth images)
  • NEO — Near Earth Object tracking and asteroid data
  • EONET — Earth Observatory Natural Event Tracker (wildfires, storms, volcanic activity)
  • DONKI — Space Weather Database of Notifications, Knowledge, and Information
  • GIBS — Global Imagery Browse Services (high-resolution Earth imagery tiles)
  • POWER — Prediction of Worldwide Energy Resources (solar radiation, wind, temperature)

jezweb/nasa-mcp-server and AnCode666/nasa-mcp offer similar coverage with different implementation approaches.

NASA APIs are free with an API key, making these servers accessible for education, research, and prototyping.

Microsoft Earth Copilot

Repository: microsoft/Earth-Copilot

While not strictly an MCP server in its original form, Earth Copilot is an AI-powered geospatial application that integrates with Microsoft Planetary Computer and NASA VEDA data catalog — providing access to 130+ satellite collections. It can be exposed as an MCP server for integration with VS Code GitHub Copilot and other MCP-compatible clients.

Platform Comparison

Server Focus Tools Language License Key Strength
GIS-MCP Core GIS operations 89 Python MIT Broadest spatial operations library
QGIS MCP Desktop GIS control 10+ Python Open Access to QGIS Processing Toolbox
ArcGIS Pro MCP Desktop GIS control Several C#/.NET 8 Open Enterprise GIS integration
CARTO MCP Cloud spatial analytics Workflow-based Managed Commercial Workflow-as-tool governance model
Mapbox MCP Location intelligence Several Node.js Commercial Real-time navigation and traffic
ArcGIS Location Location services 5+ TypeScript Open Curated global data (Esri PaaS)
GeoServer MCP Spatial data serving Several Python MIT OGC standards (WMS/WFS)
Axion Planetary Earth observation 30+ Node.js Open Google Earth Engine access
Copernicus MCP Satellite imagery Several Python Open Free ESA Sentinel data
NASA MCP Space/Earth science 20+ Various Open 20+ NASA data sources

Geospatial Foundation Models and MCP

An emerging layer in the geospatial AI stack is foundation models trained specifically on Earth observation data. These models don’t use MCP directly (yet), but they represent the analytical intelligence that MCP-connected agents can leverage:

NASA/IBM Prithvi-EO-2.0 — an open-source geospatial foundation model with 600 million parameters, six times larger than its predecessor. Co-developed by 42 members from 12 institutions, it’s pre-trained on NASA’s Harmonized Landsat and Sentinel-2 dataset. Applications include land use change detection, disaster monitoring, and crop yield prediction. Available on Hugging Face under ibm-nasa-geospatial.

As MCP adoption grows, we expect to see these foundation models exposed through MCP tool interfaces — allowing general-purpose AI agents to delegate specialized satellite image analysis to purpose-built geospatial models.

Architecture Patterns

Pattern 1: Natural Language GIS Analysis

The simplest pattern — an AI agent uses GIS-MCP or QGIS MCP to perform spatial operations requested in natural language.

User: "What's the area of overlap between these two polygons?"

┌──────────┐    ┌──────────────┐    ┌─────────────┐
│   User   │───►│   AI Agent   │───►│   GIS-MCP   │
│          │◄───│              │◄───│  (Shapely,   │
│          │    │  Interprets  │    │   PyProj,    │
│          │    │  request,    │    │   GeoPandas) │
│          │    │  formats     │    │             │
│          │    │  result      │    │  89 spatial  │
│          │    │              │    │  functions   │
└──────────┘    └──────────────┘    └─────────────┘

The agent translates the natural language request into the appropriate GIS-MCP tool calls, executes the spatial operation, and presents the result in human-readable form.

Pattern 2: Multi-Source Location Intelligence

Combining multiple geospatial MCP servers to answer complex location questions.

"Find all coffee shops within 10 minutes walk of our office,
 show them on a map, and check which ones have high ratings."

┌──────────┐    ┌──────────────┐
│   User   │───►│   AI Agent   │
│          │◄───│              │
└──────────┘    │   Chains:    │
                │              │
                │  1. Mapbox   │──► Geocode office address
                │     MCP      │──► Generate 10-min walk isochrone
                │              │
                │  2. ArcGIS   │──► Find POIs in isochrone area
                │     Location │
                │              │
                │  3. GIS-MCP  │──► Spatial filtering & ranking
                │              │
                │  4. Mapbox   │──► Render results on map
                │     MCP      │
                └──────────────┘

Pattern 3: Earth Observation Pipeline

AI agents that monitor environmental changes using satellite data.

"Monitor deforestation in this region over the past 6 months."

┌──────────┐    ┌──────────────┐
│   User   │───►│   AI Agent   │
│          │◄───│              │
└──────────┘    │   Chains:    │
                │              │
                │  1. Axion    │──► Query Earth Engine for
                │     Planetary│    Sentinel-2 imagery
                │     MCP      │──► Calculate NDVI time series
                │              │──► Run deforestation model
                │              │
                │  2. GIS-MCP  │──► Calculate area statistics
                │              │──► Generate change polygons
                │              │
                │  3. NASA MCP │──► Cross-reference EONET for
                │              │    fire events in region
                │              │
                │  Result:     │
                │  Change map, │
                │  area stats, │
                │  fire overlay│
                └──────────────┘

Pattern 4: Enterprise Spatial Decision Support

Organizations using CARTO or Mapbox to power location-aware business decisions.

"Where should we open our next store? Consider demographics,
 competition density, and drive-time accessibility."

┌──────────┐    ┌──────────────┐
│  Analyst  │───►│   AI Agent   │
│          │◄───│              │
└──────────┘    │              │
                │  1. CARTO    │──► Site selection workflow
                │     MCP      │    (demographics + competition)
                │              │
                │  2. CARTO    │──► Drive-time catchment areas
                │     MCP      │    for candidate locations
                │              │
                │  3. CARTO    │──► Revenue prediction model
                │     MCP      │    (ML on spatial features)
                │              │
                │  Result:     │
                │  Ranked site │
                │  candidates  │
                │  with maps   │
                └──────────────┘

Security and Privacy Considerations

Geospatial data carries unique security risks that don’t apply to other data types.

Location Privacy

Location data is inherently personal. Movement patterns reveal where people live, work, worship, and seek medical care. Even “anonymized” location data can be re-identified — research has shown that four spatiotemporal points are enough to uniquely identify 95% of individuals in a dataset.

Implications for MCP geospatial servers:

  • Servers handling user location data (geocoding, routing) should minimize data retention
  • Avoid logging precise coordinates in request logs where possible
  • Consider spatial generalization (rounding coordinates, using grid cells instead of exact points) for analytics

Spatial Data Integrity

GPS spoofing — broadcasting false satellite signals to trick receivers into calculating wrong positions — is a growing threat. In a geospatial MCP context, compromised location inputs could lead AI agents to make decisions based on false geography.

Implications:

  • Validate coordinate inputs against reasonable bounds
  • Cross-reference location data from multiple sources when making critical decisions
  • Be aware that user-provided coordinates may be intentionally or accidentally wrong

MCP-Specific Risks

The general MCP security risks (prompt injection, tool poisoning, token theft) apply to geospatial servers with additional spatial dimensions:

  • Prompt injection through place names — malicious instructions embedded in geocoding results, place descriptions, or feature attributes that an agent processes as tool output
  • API key exposure — geospatial services (Mapbox, Esri, Google Earth Engine) use API keys that may carry significant usage quotas or cost implications. MCP servers must handle these credentials securely
  • Data exfiltration via spatial queries — an agent tricked into executing broad spatial queries could extract sensitive location data (facility locations, infrastructure positions)
  • Schema and capability exposure — MCP tool descriptions may reveal what geospatial data an organization holds and what analyses they can perform

Best Practices

  1. Scope API keys narrowly — use separate keys for different MCP servers, with minimum required permissions
  2. Rate-limit spatial queries — prevent broad data extraction through rapid querying
  3. Audit spatial operations — log what spatial queries were executed, on what data, by which agent
  4. Validate spatial inputs — check that coordinates, bounding boxes, and geometry inputs are within expected bounds
  5. Minimize credential scope — Google Earth Engine, Esri, and Mapbox all support tiered API key permissions
  6. Handle elevation/terrain data carefully — detailed terrain data for military or critical infrastructure areas may have access restrictions

Getting Started

For Spatial Analysis Beginners

Start with GIS-MCP (mahdin75/gis-mcp). It requires no external accounts or API keys — just Python with spatial libraries. You can perform geometry operations, coordinate transformations, and spatial analysis immediately. Pair it with Claude Desktop or Cursor for a natural language GIS experience.

For QGIS Users

Install QGIS MCP (jjsantos01/qgis_mcp) to control QGIS through your AI assistant. The Processing Toolbox integration gives you access to hundreds of algorithms through natural language. This is ideal for GIS professionals who want to automate repetitive workflows.

For Earth Observation and Environmental Monitoring

Axion Planetary MCP provides the deepest satellite analysis capabilities through Google Earth Engine. For free satellite imagery without commercial restrictions, use Copernicus MCP for Sentinel data. For broader Earth science data, add a NASA MCP server.

For Location-Aware Applications

Mapbox MCP is the best starting point for applications needing geocoding, routing, and map rendering. ArcGIS Location Services MCP is an alternative if you’re in the Esri ecosystem.

For Enterprise Spatial Intelligence

CARTO MCP provides the most governed approach — workflows define exactly what spatial operations agents can perform, with full audit trails and access controls. This is the right choice for organizations that need to expose spatial analytics to AI agents while maintaining enterprise-grade governance.

What’s Next for Geospatial MCP

The intersection of MCP and geospatial is moving fast. Several trends are shaping the near future:

Real-time data integration. Mapbox predicts that by late 2026, fresh real-time data will shift from “nice-to-have” to a core requirement for agentic systems. MCP servers that connect to live traffic, weather, IoT sensors, and streaming satellite feeds will become standard.

Geospatial foundation models as MCP tools. NASA/IBM’s Prithvi-EO-2.0 and similar models will likely be wrapped in MCP servers, allowing general-purpose agents to delegate specialized Earth observation analysis to purpose-built models.

Open data standardization. The Overture Maps Foundation (backed by Amazon, Meta, Microsoft, and TomTom) is building interoperable open map data in GeoParquet format. As this dataset matures, expect MCP servers that provide structured access to Overture’s global building footprints, transportation networks, points of interest, and administrative boundaries.

Specialized geospatial agents. Rather than general-purpose agents that happen to use spatial tools, we’ll see purpose-built agents for specific domains — precision agriculture agents, disaster response agents, urban planning agents — each with curated sets of geospatial MCP tools optimized for their workflow.

Indoor spatial expansion. As indoor positioning and building information modeling (BIM) mature, MCP servers for indoor navigation, facility management, and digital twins will extend geospatial AI beyond outdoor mapping.

This guide covers the MCP geospatial ecosystem as of March 2026. Server capabilities and star counts reflect published documentation and may have changed since research was conducted. ChatForest researches and analyzes these tools — we do not deploy or test them hands-on. Have a correction or suggestion? Let us know.