mcp · Global

gfm-mcp

MCP server exposing Geospatial Foundation Model inference to LLM agents — building extraction, landcover/water segmentation, change detection, GFM embeddings. CPU-capable, 16 tools.

Visit source ↗GitHub repository ↗

Claim this listing

Represent gfm-mcp? Send a claim for review so we can discuss listing updates. Claims do not automatically grant ownership or verification.

Explore paid sponsorship →

What's in it

Data offered docs-verified: building footprint extraction (GeoJSON output) · water segmentation (NAIP / Sentinel-2 bands configurable) · object detection (solar_panel, ship, car) · bi-temporal change detection (ChangeStar) · zero-shot scene classification (CLIP) · GFM embeddings (tessera 128-d, GSE/AlphaEarth 64-d, galileo 192-d) · embedding similarity and K-means embedding clustering · GFM model catalogue (list_gfm_models, get_model_info)
Coverage llm-inferred: Global
Formats docs-verified: GeoJSON · MCP stdio
Access docs-verified: Open source, local install (Python 3.12+). Licence GPL-3.0-or-later per DOCS (ENTRY says GPL-3.0). GEE-backed embedding tools need Google OAuth, non-commercial registration and project binding.

Useful for

Use cases llm-inferred: Building footprint extraction from aerial imagery as exposure or drainage catchment inputs · Water-body segmentation from satellite or aerial imagery for flood extent baseline mapping · Bi-temporal change detection for land-use or built-form change screening across survey dates · GFM embedding similarity and clustering to zone a site or study area by surface form
Audience llm-inferred: GIS analysts · remote sensing analysts · flood consultants · hydrologists · ecologists · planners

Limitations

Constraints docs-verified: Embedding tools (tessera, gse, galileo) need a proxy to reach Google-hosted data; without one they return a structured error rather than results · GEE-backed embeddings require Google OAuth, non-commercial registration and project binding · On-the-fly embeddings for other models (e.g. prithvi) are an unchecked roadmap item · finetune_remote (TerraTorch via Colab) is roadmap only and is not among the 16 tools · Verified only on Windows 11 x64, CPU-only; other platforms and GPU use are not documented · Upstream geoai 0.42.0 masks_to_vector regularize bug is worked around by fixing regularize=False · Heavy tests need a sample imagery file, and model weights download on first use · Pre-1.0 version (0.2.x) with heavy upstream dependencies (geoai-py, torch, rs-embed)

Licence: GPL-3.0
Authentication: OAuth mentioned in README (auto-extracted)
GitHub stars: 3
Last commit: 29/08/2026
Last successful page check: 14/10/2026
Page health: healthy
Package: Not recorded
Verification: Not independently verified

Auto-extracted, unverified fields: tools, install, auth. Check upstream documentation.

Provenance & evidence

Tools/install/auth auto-extracted from README; quality score heuristic (stars, recency, tool count, licence).

Metadata observed 15/10/2026 · research sweep: github-search