List server capabilities: catalogs, collections, and band info.

Returns a comprehensive overview of what this server can do, including supported catalogs, satellite collections, band names per platform, and available spectral indices with required bands.

Args: output_mode: Response format - "json" (default) or "text"

Returns: JSON with catalogs, collections, band mappings, and spectral indices

Tips for LLMs: - Call this FIRST to understand what the server can do before planning any analysis workflow - Use band_mappings to know which band names to pass to download tools - Use spectral_indices to see which indices are available and what bands they require - Typical workflow: stac_capabilities → stac_search → stac_describe_scene → stac_download_bands or stac_compute_index - Collections: sentinel-2-l2a (optical, 10m), landsat-c2-l2 (optical, 30m), sentinel-1-grd (SAR radar, 10m), cop-dem-glo-30 (elevation, 30m)

Example: caps = await stac_capabilities()

Compute a spectral index (e.g., NDVI, NDWI) for a scene.

Automatically downloads the required bands, computes the index formula, and stores the result as a single-band float32 raster.

Supported indices:

• ndvi: Vegetation (NIR - Red) / (NIR + Red)

• ndwi: Water (Green - NIR) / (Green + NIR)

• ndbi: Built-up (SWIR16 - NIR) / (SWIR16 + NIR)

• evi: Enhanced Vegetation 2.5*(NIR - Red) / (NIR + 6Red - 7.5Blue + 1)

• savi: Soil-Adjusted Vegetation ((NIR - Red) / (NIR + Red + 0.5)) * 1.5

• bsi: Bare Soil ((SWIR16 + Red) - (NIR + Blue)) / ((SWIR16 + Red) + (NIR + Blue))

Args: scene_id: Scene identifier from a previous search index_name: Index to compute (ndvi, ndwi, ndbi, evi, savi, bsi) bbox: Optional crop bbox in EPSG:4326 [west, south, east, north] cloud_mask: Apply SCL-based cloud masking before computation (Sentinel-2 only) output_format: Output format - "geotiff" (default) or "png" output_mode: Response format - "json" (default) or "text"

Returns: JSON with artifact_ref and value_range for the computed index

Tips for LLMs: - Use stac_capabilities to see all available indices and required bands - Interpretation guide: - NDVI: >0.6 dense vegetation, 0.2-0.6 moderate, <0.2 bare/water - NDWI: >0 water, <0 land; useful for flood mapping - NDBI: >0 built-up, <0 natural land cover - EVI: similar to NDVI but corrects for atmospheric and soil effects - SAVI: like NDVI but better in areas with sparse vegetation - BSI: >0 bare soil, <0 vegetated - Only works for optical collections (Sentinel-2, Landsat) - Enable cloud_mask=True for cleaner results with Sentinel-2 data - For temporal change analysis, compute the same index on multiple dates and compare value_range

Example: ndvi = await stac_compute_index( scene_id="S2B_...", index_name="ndvi", bbox=[0.85, 51.85, 0.95, 51.92] )

Check if cached scenes fully cover a requested bounding box.

Rasterizes the target bbox into a grid and checks which cells are covered by the provided scenes. Useful for planning mosaics to ensure gap-free coverage.

Args: bbox: Target bounding box [west, south, east, north] in EPSG:4326 scene_ids: Scene identifiers (must be cached from prior stac_search calls) output_mode: Response format - "json" (default) or "text"

Returns: JSON with coverage percentage and uncovered areas

Tips for LLMs: - Call this before stac_mosaic to verify scenes fully cover your area of interest - If coverage is less than 100%, search for more scenes or widen the date range to find additional tiles - Scenes must have been found by a prior stac_search call

Example: check = await stac_coverage_check( bbox=[0.8, 51.8, 1.0, 51.95], scene_ids=["S2B_...", "S2A_..."] )

Get detailed information about a STAC collection.

Returns band wavelengths, recommended composites, supported spectral indices, cloud masking info, and LLM-friendly usage guidance.

For known collections (Sentinel-2, Landsat, Sentinel-1, DEM), provides rich metadata including band names needed for download tools. Unknown collections still return live STAC metadata.

Args: collection_id: Collection identifier. Options: sentinel-2-l2a, sentinel-2-c1-l2a, landsat-c2-l2, sentinel-1-grd, cop-dem-glo-30 catalog: Catalog name (default: earth_search). Options: earth_search, planetary_computer, usgs output_mode: Response format - "json" (default) or "text"

Returns: JSON with band details, composites, spectral indices, and guidance

Tips for LLMs: - Call this to discover band names before using stac_download_bands - The composites field lists pre-defined band combinations (e.g., true_color = [red, green, blue]) - The spectral_indices field shows which indices this collection supports - The llm_guidance field contains domain-specific usage advice - Check cloud_mask_band — if None, the collection is non-optical (SAR radar or DEM) and cloud_mask=True will fail

Example: detail = await stac_describe_collection( collection_id="sentinel-2-l2a" )

Get detailed information about a specific scene.

Shows all available assets/bands, properties, CRS, and download URLs. The scene must have been returned by a previous stac_search call.

Args: scene_id: Scene identifier from a search result (use scene_id from stac_search) output_mode: Response format - "json" (default) or "text"

Returns: JSON with full scene details including all assets, CRS, and properties

Tips for LLMs: - Call this after stac_search to see available bands before downloading - The assets list shows every downloadable band and its resolution - Use the band keys (e.g., red, nir, scl) in stac_download_bands - Check cloud_cover to decide if the scene is usable

Example: detail = await stac_describe_scene( scene_id="S2B_MSIL2A_20240715T105629_N0510_R094_T31UCR_20240715T143301" )

Download specific bands from a scene as a GeoTIFF or PNG.

Reads band COGs via HTTP, windows to the requested bbox, and stores the result in chuk-artifacts. The bbox should be in EPSG:4326 — CRS reprojection to the raster's native CRS is handled automatically.

Args: scene_id: Scene identifier from a previous stac_search call bands: Band names to download. Common names: - Sentinel-2: red, green, blue, nir, swir16, swir22, rededge1-3, scl - Landsat: red, green, blue, nir08, swir16, swir22, coastal, qa_pixel - Sentinel-1: vv, vh - DEM: data bbox: Optional crop bbox in EPSG:4326 [west, south, east, north]. Strongly recommended to avoid downloading full tiles output_format: "geotiff" (default, lossless, for analysis) or "png" (8-bit lossy with percentile stretch, for preview/display) cloud_mask: Apply SCL-based cloud masking (Sentinel-2 only). Masked pixels become 0 (integer) or NaN (float) output_mode: Response format - "json" (default) or "text"

Returns: JSON with artifact_ref, shape, dtype, and optional preview_ref

Tips for LLMs: - Use stac_describe_scene first to see available band names - Always provide a bbox to limit download size - Use output_format="png" when the user wants to see the image - GeoTIFF preserves full radiometric precision for analysis - A PNG preview is auto-generated alongside GeoTIFF downloads - For RGB visualisation, prefer stac_download_rgb (simpler) - For spectral indices, prefer stac_compute_index (automatic)

Example: result = await stac_download_bands( scene_id="S2B_...", bands=["red", "nir"], bbox=[0.85, 51.85, 0.95, 51.92] )

Download a multi-band composite from a scene.

Creates a composite from any combination of bands. Band order determines RGB channel mapping (first=R, second=G, third=B).

Args: scene_id: Scene identifier from a previous stac_search call bands: Band names for the composite (order = R,G,B channels). Common recipes: - ["nir", "red", "green"] — false colour infrared (vegetation=red) - ["swir16", "nir", "red"] — agriculture (crops=bright green) - ["swir16", "swir22", "red"] — geology/minerals composite_name: Label for the composite (e.g., "false_color_ir") bbox: Optional crop bbox in EPSG:4326 [west, south, east, north] output_format: "geotiff" (default, lossless) or "png" (8-bit preview) cloud_mask: Apply SCL-based cloud masking (Sentinel-2 only) output_mode: Response format - "json" (default) or "text"

Returns: JSON with artifact_ref for the composite

Tips for LLMs: - Use stac_describe_collection to see pre-defined composite recipes - Band order matters: first band → Red, second → Green, third → Blue - For true-colour RGB, use stac_download_rgb instead (simpler) - For single-value analysis, use stac_compute_index (e.g., NDVI)

Example: false_color = await stac_download_composite( scene_id="S2B_...", bands=["nir", "red", "green"], composite_name="false_color_ir" )

Download a true-color RGB composite from a scene.

Convenience wrapper around stac_download_bands that automatically selects red, green, blue bands. This is the simplest way to get a visual satellite image.

Args: scene_id: Scene identifier from a previous stac_search call bbox: Optional crop bbox in EPSG:4326 [west, south, east, north]. Strongly recommended to avoid downloading full tiles output_format: "geotiff" (default, lossless) or "png" (8-bit lossy, suitable for inline display by LLMs) cloud_mask: Apply SCL-based cloud masking (Sentinel-2 only) output_mode: Response format - "json" (default) or "text"

Returns: JSON with artifact_ref for the RGB composite

Tips for LLMs: - Use output_format="png" when the user wants to see the image — PNG can be rendered inline - GeoTIFF preserves full 16-bit precision but cannot be displayed inline - For false-color composites (e.g., NIR/Red/Green), use stac_download_composite instead - Only works for optical collections (Sentinel-2, Landsat) that have red, green, blue bands

Example: rgb = await stac_download_rgb(scene_id="S2B_...", output_format="png")

Estimate download size for bands from a scene (no pixel data read).

Reads only COG headers to determine dimensions, dtype, and estimated file size. Use this before large downloads to understand how much data will be transferred.

Args: scene_id: Scene identifier from a previous search bands: Band names to estimate (e.g., ["red", "green", "blue", "nir"]) bbox: Optional crop bbox in EPSG:4326 [west, south, east, north] output_mode: Response format - "json" (default) or "text"

Returns: JSON with per-band size details and total estimate

Tips for LLMs: - Call this BEFORE large downloads to check feasibility - If estimated_mb > 500, suggest a smaller bbox or fewer bands - No pixel data is read — only COG headers, so this is very fast - Use the per-band breakdown to see which bands are largest (e.g., 10m bands are ~4x larger than 20m bands) - Useful for planning stac_mosaic or stac_temporal_composite where multiple scenes multiply the total data volume

Example: estimate = await stac_estimate_size( scene_id="S2B_...", bands=["red", "nir"], bbox=[0.85, 51.85, 0.95, 51.92] )

Find before/after scene pairs for change detection.

Searches two date ranges and matches scenes by spatial overlap, useful for detecting changes between time periods (e.g., flood damage, urban growth, deforestation, seasonal vegetation change).

Args: bbox: Bounding box [west, south, east, north] in EPSG:4326 before_range: Before date range "YYYY-MM-DD/YYYY-MM-DD" after_range: After date range "YYYY-MM-DD/YYYY-MM-DD" collection: STAC collection (default: sentinel-2-l2a). Options: sentinel-2-l2a, sentinel-2-c1-l2a, landsat-c2-l2, sentinel-1-grd, cop-dem-glo-30 max_cloud_cover: Maximum cloud cover percentage 0-100 (default: 20). Ignored for non-optical collections (sentinel-1-grd, cop-dem-glo-30). catalog: Catalog name (default: earth_search). Options: earth_search, planetary_computer, usgs output_mode: Response format - "json" (default) or "text"

Returns: JSON with matched scene pairs sorted by overlap percentage

Tips for LLMs: - Best for change detection workflows: find pairs, then download the same bands for before/after scenes and compare - For flood mapping: use sentinel-1-grd (SAR sees through clouds) with before_range = dry season, after_range = flood event - For vegetation change: use sentinel-2-l2a, then compute NDVI for each scene in the pair - Higher overlap_percent means better spatial coverage for comparison - Follow up with stac_download_bands or stac_compute_index on each scene in the pair

Example: pairs = await stac_find_pairs( bbox=[0.8, 51.8, 1.0, 51.95], before_range="2024-01-01/2024-03-31", after_range="2024-07-01/2024-09-30" )

Retrieve a stored artifact and save it as a local file for viewing.

Downloads the artifact bytes from the artifact store and writes them to a local file. Returns the file path so the image can be opened in any viewer.

Args: artifact_ref: Artifact ID from a previous download tool call (the artifact_ref or preview_ref value) output_mode: Response format - "json" (default) or "text"

Returns: JSON with file_path, mime type, size, and artifact metadata

Tips for LLMs: - Use the preview_ref (PNG) from download results for quick viewing - Use the artifact_ref (GeoTIFF) for full-resolution geospatial data - The file is saved to a temporary directory and can be opened with any image viewer or GIS application - PNG files can be opened with: open <file_path> (macOS) - GeoTIFF files can be opened with QGIS, rasterio, or similar

Example: result = await stac_get_artifact( artifact_ref="a7c666e6555548aea2d5351cc65bf173" ) # Returns: {"file_path": "/tmp/stac_artifacts/a7c666...png", ...}

Check which STAC API features a catalog supports.

Reads the catalog's conformance URIs and matches them against known STAC API conformance classes to determine feature support (core, item_search, filter, sort, fields, query, collections).

Args: catalog: Catalog name (default: earth_search). Options: earth_search, planetary_computer, usgs output_mode: Response format - "json" (default) or "text"

Returns: JSON with feature support flags and raw conformance URIs

Tips for LLMs: - Rarely needed — most workflows don't require conformance checking - Useful for debugging when a catalog doesn't support expected features - Check for "query" support if advanced filtering is needed

Example: conformance = await stac_get_conformance(catalog="earth_search")

List all known STAC catalogs.

Returns pre-configured catalog endpoints that can be searched. Each catalog hosts different satellite collections.

Args: output_mode: Response format - "json" (default) or "text"

Returns: JSON with catalog names and endpoint URLs

Tips for LLMs: - Use stac_capabilities instead for a full overview including collections, bands, and indices - Default catalog is earth_search (AWS Element 84) - planetary_computer requires no API key (auto-authenticated)

Example: catalogs = await stac_list_catalogs()

List available collections in a STAC catalog.

Queries the catalog's live API to discover all hosted collections with titles, descriptions, and spatial/temporal extents.

Args: catalog: Catalog name (default: earth_search). Options: earth_search, planetary_computer, usgs output_mode: Response format - "json" (default) or "text"

Returns: JSON list of collections with titles, descriptions, and extents

Tips for LLMs: - Use this to discover what data is available in a specific catalog - For detailed band/composite info on a collection, follow up with stac_describe_collection - Different catalogs host different collections — if a collection isn't found, try another catalog

Example: collections = await stac_list_collections(catalog="earth_search")

Visualise STAC scene search results as a multi-layer footprint map. Scenes appear as bounding-box polygons, grouped by collection, with cloud cover, acquisition date, and thumbnail URL in the popup. Run stac_search first, then pass the scene_ids here.

Create a mosaic from multiple scenes.

Combines overlapping scenes into a single seamless raster. Useful when your area of interest spans multiple satellite tiles.

Args: scene_ids: List of scene identifiers to mosaic (from stac_search) bands: Bands to include (e.g., ["red", "green", "blue"]) bbox: Output bounding box [west, south, east, north] in EPSG:4326. Defaults to union of all scenes if not specified output_format: "geotiff" (default, lossless) or "png" (8-bit preview) cloud_mask: Apply SCL-based cloud masking per scene before merge (Sentinel-2 only) method: Merge method: - "last" (default): later scenes overwrite earlier in overlap areas - "quality": SCL-based best-pixel selection — picks the clearest pixel from overlapping scenes (Sentinel-2 only) output_mode: Response format - "json" (default) or "text"

Returns: JSON with artifact_ref for the mosaic raster

Tips for LLMs: - Use stac_coverage_check first to verify scenes cover the target area - Use method="quality" for cloud-free mosaics from Sentinel-2 data - Use method="last" for quick mosaics or non-optical data - For temporal compositing (e.g., seasonal median), use stac_temporal_composite instead

Example: mosaic = await stac_mosaic( scene_ids=["S2B_001", "S2B_002"], bands=["red", "green", "blue"], method="quality" )

Visualise before/after scene pairs from stac_find_pairs as a two-layer map. Blue = before, red = after. Toggle layers to compare footprint coverage between time periods.

Get a preview/thumbnail URL for a scene.

Returns the URL of the scene's thumbnail or rendered preview image. Much faster than downloading full bands — useful for quick browsing.

The scene must have been returned by a previous stac_search call.

Args: scene_id: Scene identifier from a search result (use scene_id from stac_search) output_mode: Response format - "json" (default) or "text"

Returns: JSON with preview_url for the scene's thumbnail

Tips for LLMs: - Use for quick visual checks before committing to full band downloads - The preview URL is a remote image that can be displayed directly - Not all scenes have thumbnails — check for errors in the response

Example: preview = await stac_preview(scene_id="S2B_...")

Fetch queryable properties from a STAC API.

Returns the properties that can be used in search queries, including their types and allowed values. Useful for understanding what filtering options are available.

Args: catalog: Catalog name (default: earth_search). Options: earth_search, planetary_computer, usgs collection: Optional collection to scope queryables (e.g., "sentinel-2-l2a" for collection-specific properties) output_mode: Response format - "json" (default) or "text"

Returns: JSON with queryable property names, types, and descriptions

Tips for LLMs: - Rarely needed — stac_search handles common filters automatically - Useful when debugging why searches return unexpected results - Different catalogs support different queryable properties

Example: queryables = await stac_queryables( catalog="earth_search", collection="sentinel-2-l2a" )

Search for satellite scenes matching spatial and temporal criteria.

This is the primary entry point for finding satellite imagery. Results are cached for follow-up describe/download operations.

Args: bbox: Bounding box [west, south, east, north] in EPSG:4326. Example: [-0.7, 52.7, 0.5, 53.7] for Lincolnshire, UK collection: STAC collection (default: sentinel-2-l2a). Options: - sentinel-2-l2a: Optical imagery, 10-20m resolution, 13 bands - sentinel-2-c1-l2a: Reprocessed Sentinel-2 archive - landsat-c2-l2: Optical imagery, 30m resolution, 11 bands - sentinel-1-grd: SAR radar, 10m, sees through clouds (VV/VH) - cop-dem-glo-30: Global elevation data, 30m date_range: Date range as "YYYY-MM-DD/YYYY-MM-DD" (optional). Omit for cop-dem-glo-30 (no temporal dimension) max_cloud_cover: Maximum cloud cover percentage 0-100 (default: 20). Ignored for non-optical collections (sentinel-1-grd, cop-dem-glo-30). Increase to 30-50 if getting zero results max_items: Maximum results to return (default: 10) catalog: Catalog name (default: earth_search). Options: earth_search, planetary_computer, usgs output_mode: Response format - "json" (default) or "text"

Returns: JSON with matching scenes sorted by cloud cover (optical) or date

Tips for LLMs: - Typical workflow: stac_search → stac_describe_scene → stac_download_bands - For cloudy regions (e.g., UK autumn), increase max_cloud_cover to 50 or use sentinel-1-grd (SAR radar, not affected by clouds) - For flood mapping: use sentinel-1-grd (water appears dark in VV/VH) - For vegetation analysis: use sentinel-2-l2a with NDVI index - For elevation/terrain: use cop-dem-glo-30 (no date_range needed) - If zero results, check the hints field for suggestions (try different catalog, increase cloud cover, widen date range) - Scenes are cached — use scene_id in subsequent describe/download calls

Example: results = await stac_search( bbox=[0.8, 51.8, 1.0, 51.95], date_range="2024-06-01/2024-08-31", max_cloud_cover=10 )

Get server status and configuration.

Returns information about the server version, available catalogs, and artifact store status. Use this to verify the server is running and check storage configuration.

Args: output_mode: Response format - "json" (default) or "text"

Returns: JSON with server status including version, storage provider, and default catalog

Tips for LLMs: - Call stac_capabilities instead if you need to plan a workflow (it returns collections, bands, and indices too) - Check artifact_store_available before attempting downloads

Example: status = await stac_status()

Create a temporal composite by combining multiple scenes statistically.

Searches for scenes in the date range, downloads bands from each, then combines them pixel-by-pixel using a statistical method. Useful for creating cloud-free composites from cloudy time series.

Args: bbox: Area of interest [west, south, east, north] in EPSG:4326 bands: Bands to composite (e.g., ["red", "green", "blue"]) date_range: Date range "YYYY-MM-DD/YYYY-MM-DD" method: Statistical method - "median" (default), "mean", "max", "min" collection: STAC collection (default: sentinel-2-l2a) max_cloud_cover: Maximum cloud cover 0-100 (default: 20) max_items: Maximum scenes (default: 10) catalog: Catalog name (default: earth_search) cloud_mask: Apply SCL cloud masking per scene before compositing output_format: Output format - "geotiff" (default) or "png" output_mode: Response format - "json" (default) or "text"

Returns: JSON with artifact_ref for the temporal composite

Tips for LLMs: - Method selection: - "median" (default): best for cloud-free composites — robust to outliers from clouds/shadows - "mean": smooth average, good for general baselines - "max": captures peak values (e.g., peak NDVI in a growing season) - "min": captures minimum values (e.g., lowest water extent) - Enable cloud_mask=True with Sentinel-2 for best results — masks clouds before compositing so they don't affect the statistics - Use a 2-3 month date range for seasonal composites - For per-date outputs instead of a single composite, use stac_time_series - Cloud cover filter is automatically skipped for non-optical collections (sentinel-1-grd, cop-dem-glo-30) - max_items defaults to 10 — increase for denser temporal sampling

Example: composite = await stac_temporal_composite( bbox=[0.85, 51.85, 0.95, 51.92], bands=["red", "green", "blue"], date_range="2024-06-01/2024-08-31", method="median" )

Extract a time series of band data over an area.

Searches for all scenes in the date range, downloads the requested bands for each, and returns references to the full temporal stack.

Args: bbox: Area of interest [west, south, east, north] bands: Bands to extract (e.g., ["red", "nir"]) date_range: Date range "YYYY-MM-DD/YYYY-MM-DD" collection: STAC collection (default: sentinel-2-l2a) max_cloud_cover: Maximum cloud cover 0-100 (default: 20) max_items: Maximum scenes to include (default: 50) catalog: Catalog name (default: earth_search) output_mode: Response format - "json" (default) or "text"

Returns: JSON with per-date artifact references

Tips for LLMs: - Use for monitoring change over time (vegetation growth, flood extent, urban expansion) - Pair with stac_compute_index on each date's artifact for temporal index analysis (e.g., NDVI over a growing season) - Keep bbox small — each date downloads full band data - Use max_cloud_cover=10 for cleaner optical time series - For a single cloud-free image from a date range, use stac_temporal_composite with method="median" instead - max_items limits the number of dates; set higher for dense temporal sampling or lower to reduce download volume - Cloud cover filter is automatically skipped for non-optical collections (sentinel-1-grd, cop-dem-glo-30)

Example: ts = await stac_time_series( bbox=[0.85, 51.85, 0.95, 51.92], bands=["red", "nir"], date_range="2024-01-01/2024-12-31", max_cloud_cover=10 )

Read out a raster's values within zones — the inference step after a fetch.

Given a stored raster artifact (e.g. an NDVI index from stac_compute_index, or any GeoTIFF from stac_download_bands) and target zones, return per-zone statistics (n_valid, mean, std, min, max, median, p10, p90).

Zones are either circular buffers around points, or GeoJSON polygons:

• points: [[x, y], ...] centres in zones_crs, each summarised within buffer_m.

• geojson: a geometry / Feature / FeatureCollection (polygons) in zones_crs.

Pass background_m (> buffer_m) to also get a LOCAL ANOMALY readout: each point's mean is compared to the surrounding annulus (buffer_m..background_m) and reported as a z-score, with anomalous set when |z| >= z_threshold. This is the direct "is the signal at this location anomalous vs its surroundings?" answer — e.g. a cropmark/soil-mark over a buried feature in an NDVI raster.

Args: artifact_id: A GeoTIFF raster artifact (NOT a PNG preview). points: Zone centres [[x, y], ...] in zones_crs (e.g. BNG eastings/northings). buffer_m: Circular zone radius in metres (raster must be projected). background_m: Outer annulus radius in metres → enables the z-score readout. geojson: Alternative polygon zones (geometry/Feature/FeatureCollection). zones_crs: CRS of points/geojson, e.g. "EPSG:27700" (BNG) or "EPSG:4326". labels: Optional labels for points (e.g. HER refs). band: 1-based band index to read. z_threshold: |z| at/above which a point is flagged anomalous (default 2.0). output_mode: "json" or "text".

Returns: Per-zone statistics, plus a local z-score when background_m is given.