mcp · Global

glen-sdm

Species distribution models on the H3 grid, served over MCP

Claim this listing

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

Data offered docs-verified: GBIF occurrence records (2026-09 snapshot, quality-filtered) · user-supplied presence points (GeoJSON or coordinate list) · CHELSA 2.1 bioclim (7 variables) · Copernicus DEM elevation and roughness · CGLS-LC100 land-cover fractions (8 groups) · Global Human Modification · user-named H3 parquet covariates · ensemble suitability surface on H3 grid · binary range polygon · spatial CV metrics · optional future projection (gained/lost area, extrapolation flags)
Coverage docs-verified: Study area chosen per run: default buffer around occurrence records, ISO region or country (Overture), or bounding box
Resolution docs-verified: H3 res 5–8 model cells (finest that keeps domain under 2 million cells); covariates at res 8–9; suitability output coarsened to at most 40,000 cells
Formats docs-verified: Parquet (predictions.parquet) · GeoJSON (range.geojson, map_layers) · JSON (summary.json) · H3 hex tiles via register_hex_tiles SQL · MCP streamable HTTP
Access llm-inferred: Self-hosted MCP server (Python; run locally with python server.py or deployed on Kubernetes via k8s/ manifests). No authentication or licence stated in docs; underlying data is read from a public S3 STAC catalog

Use cases llm-inferred: Screening a protected species' likely range and suitability surface for a site-search or constraints study · Building a species distribution baseline from user-supplied survey points (GeoJSON) for a habitat assessment · Projecting habitat gain and loss under a CHELSA future climate scenario for climate-risk appraisal
Audience llm-inferred: ecologists · ecological consultants · GIS analysts · SDM modellers · conservation planners

Constraints llm-inferred: Marked as a first pass for review by SDM experts, not yet validated as production-grade · TSS and AUC are computed against background cells, not true absences · No collinearity screening of covariates · Extrapolation flag is a count of out-of-range covariates, not a full MESS · Results and job state stored on one pod's disk unless S3 output is configured · Continental-scale runs are slow because res-8 covariates are read and averaged up · Docs do not describe UK-specific covariates or validation · Requires self-hosting or a reachable server; no auth documented

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

Daily discovery 2026-10-16: GitHub search; verified via GitHub API. Tools/install/auth auto-extracted from README.

Metadata observed 14/10/2026 · daily-2026-10-16 (github-search)