AI Agent MCP Servers • RustGIS WASM • OGC APIs • On-Demand GPU GIS

AI-Ready Geospatial MCP Servers, Web Portals & Pro Workstations

Modern geospatial platforms must serve both human analysts and autonomous AI agents. Tech Maven Geospatial pairs catalog-governed Model Context Protocol (MCP) servers with a client-side RustGIS WebAssembly Geoportal, a Serverless Operations Dashboard, standards-based OGC / STAC APIs, and on-demand NVIDIA GPU desktop GIS workstations.

Explore Dual-Path GeoQuery MCP RustGIS WASM & Ops Dashboard On-Demand ArcGIS Pro & QGIS

Dual-Path MCP

<1s PostGIS Lambda Fast Path + Disposable DuckDB Batch Lakehouse Workers

Zero Raw SQL

AI agents invoke approved catalog IDs & bound parameters—never arbitrary SQL/URLs

WASM + OGC

RustGIS browser rendering, pgSTAC discovery & WMS/WFS/WCS/OGC Features

G4dn GPU GIS

Stopped-by-default Windows EC2 with NVIDIA T4 for ArcGIS Pro & QGIS

Model Context Protocol (MCP) Architecture

Safe, Auditable Geospatial Intelligence for LLMs & AI Agents

Strict separation between AI tool calls and underlying database engines eliminates SQL injection, credential leakage, and unbounded scan costs.

1. GeoQuery MCP Server (`geoquery/`)

Dual-Engine Spatial Question Answering

Allows AI assistants (Claude Desktop, Amazon Bedrock Agents, Cursor, custom MCP clients) to answer complex spatial questions using approved catalog datasets and validated templates.

Synchronous PostGIS Fast Path (`geoquery/postgis_fastpath.py`): Routes low-latency 2D vector, 3D volumetric (`spatial3d`), statistical (`geostats`), suitability (`geosuitability`), and DEM raster (`georaster`) questions directly to the serverless PostGIS Lambda microservices in under a second.
Asynchronous DuckDB 2.0 Batch Path: Automatically dispatches large-area GeoParquet/Iceberg analytical templates to disposable Fargate DuckDB workers with bounded output rows and S3 result presigning.
Zero Arbitrary SQL or URLs: MCP requests must reference a registered `dataset_id` and `template_id` from `config/geoquery/catalog.json` with strongly typed bound parameters (`bbox`, `radius_m`, `z_max`, etc.).

2. GeoAGENTS Platform Admin MCP (`geoagents_mcp/`)

Conversational DevOps & Pipeline Orchestration

Empowers platform engineers and GIS administrators to inspect infrastructure health, trigger batch runs, and manage spatial catalogs through natural language.

Batch & Step Functions Control: Query active AWS Batch queues, inspect job logs, and submit parameterized raster/DuckDB/Terrio runs with human-in-the-loop confirmation.
DynamoDB Lease & Run Auditing: Check monotonic latest-run pointers, release stale HRRR/pipeline leases, and verify atomic S3 output markers.
Scoped Connector Secrets: Uses AWS Secrets Manager references (`secret_arn`) rather than passing credentials in agent context windows.

Interactive Visualization & Interoperability

WebAssembly Geoportal, Serverless Ops Console & OGC Publishing

Zero-install browser visualization paired with enterprise OGC and STAC interoperability.

RustGIS Fork (`rustgis-fork/`)

Rust + WebAssembly Geoportal

Compiles high-performance Rust spatial algorithms to WebAssembly (WASM) so users can inspect Cloud-Optimized GeoTIFFs, FlatGeobuf, GeoParquet, and 3D extruded buildings directly in the browser while offloading heavy operations to AWS Lambda & Batch.

Ready-to-go mapping apps can also be hosted as static sites on S3. They are powered by WASM and advanced JS/TS, and they work with cloud-native and cloud-optimized formats.

Ops Console (`dashboard/`)

Serverless Geospatial Ops Dashboard

A modern dark-slate operations console for monitoring AWS Batch Fargate/EC2 queues, invoking PostGIS Lambda microservices, browsing STAC catalog collections, and inspecting live CloudWatch job telemetry.

OGC & STAC (`infra/neoserver-ecs.yaml`)

Neoserver OGC & pgSTAC Catalog

Publishes S3 COGs, Zarr cubes, and PostGIS layers as standard WMS, WFS, WCS, WMTS, and OGC API - Features / Tiles endpoints alongside a SpatioTemporal Asset Catalog (pgSTAC) for instant discovery in QGIS, ArcGIS, and Leaflet/MapLibre.

On-Demand Cloud Workstations (`infra/arcgis-workstation.yaml`)

Elastic NVIDIA GPU Workstations for ArcGIS Pro & QGIS

Keep desktop GIS right next to your S3 data lake and Aurora PostGIS database—paying for EC2 GPU compute only during active editing sessions.

Zero Latency Between Desktop GIS and Your Cloud Lakehouse

Downloading 200 GB of imagery or LiDAR over office broadband to run desktop GIS kills productivity. Tech Maven Geospatial provisions GPU-accelerated Amazon EC2 `g4dn.xlarge` / `g4dn.2xlarge` (NVIDIA T4) workstations inside the same AWS region as your S3 buckets and Aurora PostGIS cluster.

Stopped by Default & Auto-Stop Idle Alarms: Instances remain stopped (`$0/hr` compute cost) until started by an analyst and automatically hibernate/stop when CloudWatch CPU/network utilization drops below threshold.
High-Bandwidth S3 Mountpoint & PostGIS Access: Stream Cloud-Optimized GeoTIFFs, COPC LiDAR, and GeoParquet at multi-gigabit VPC speeds directly inside ArcGIS Pro and QGIS.
NICE DCV / Fleet Manager Streaming: Pixel-accurate 60 FPS remote 3D globe and stereo photogrammetry streaming over HTTPS.

Workstation CloudFormation Highlights (`infra/arcgis-workstation.yaml`)

Resources:
  GisWorkstationInstance:
    Type: AWS::EC2::Instance
    Properties:
      InstanceType: g4dn.xlarge # NVIDIA T4 16GB GPU
      IamInstanceProfile: !Ref GisWorkstationProfile
      BlockDeviceMappings:
        - DeviceName: /dev/sda1
          Ebs:
            VolumeType: gp3
            VolumeSize: 250
            Throughput: 500
            Encrypted: true
  IdleAutoStopAlarm:
    Type: AWS::CloudWatch::Alarm
    Properties:
      AlarmActions:
        - !Sub "arn:aws:automate:${AWS::Region}:ec2:stop"