AgenticGeoMate
Australian geoscience data for agents: exploration reports, drillholes, geochemistry, WA mineral prospectivity and read-only SQL. Remote server with OAuth sign-in (free account). Built by AI4WA and UWA NTLP.
託管 MCP 伺服器
npx add-mcp 'https://geomate.nlp-tlp.org/mcp'可安裝到 Claude Code、Codex、Cursor 等客戶端
文件
One REST API over the whole database. Sign in, get a token, then follow the guides below; the full reference is at the end. GIS users can add the public layers straight from the OGC service at /ogc.
Quick start
Create a personal token on your account page, install the Python client, ask. Or send the token as a bearer from anything that speaks HTTP.
pip install "git+https://github.com/PascalSun/[email protected]"
from ntlp_geodata import GeoData
gd = GeoData(token="gp_your_token_here")
print(gd.nearby(121.5, -30.7, radius_km=10))
curl "https://platform.nlp-tlp.org/geo/api/nearby/?lon=121.5&lat=-30.7" -H "Authorization: Bearer <token>"
Agents
The same data as MCP tools, for Claude, Cursor or your own agent. Sign in with a platform account when the client asks.
https://geomate.nlp-tlp.org/mcp
Getting the data, step by step
Start at the top and each guide hands you what the next one needs. Get data with the Python client; build your own site with the web client.
[
Python client — get the data ↗︎
For notebooks and scripts: every guide below, paging and rate limits handled.
pip install "git+https://github.com/PascalSun/[email protected]"](https://github.com/PascalSun/ntlp-geodata)[
Web client — build a site on the platform ↗︎
For your own web app: the platform’s sign-in and authenticated calls, as MineTrace and CoreFlagging use them.
"@ntlp/geodata-web": "github:PascalSun/ntlp-geodata-web"](https://github.com/PascalSun/ntlp-geodata-web)
- 01
Sign in and get a token
Everything but the summary and the tiles needs a token. For scripts and notebooks, create a personal token on your account page: Read covers every query, Read and write also uploads and edits your data, and it can expire after 30, 90 or 365 days. For an app that signs people in, use the sign-in code and refresh flow. With a token you may make 600 requests a minute (SQL 30, exports 10 an hour); past that the answer is 429 with a Retry-After.
Now you have: A bearer token for every call below, and a refresh token to renew it.pip install "git+https://github.com/PascalSun/[email protected]" from ntlp_geodata import GeoData gd = GeoData(token="gp_your_token_here") # a personal token from https://geo.nlp-tlp.org/map?panel=account#tokens gd.me() # or set GEODATA_TOKEN and call GeoData() # every call below follows pages for you and waits out a 429 - 02
See what is there
Before pulling anything, see how much the platform holds and which datasets exist, with their licences.
Now you have: The counts you see on the front page, and the dataset list behind https://geo.nlp-tlp.org/data.gd.summary() # no token needed gd.catalog(q="drill") # the data catalogue behind https://geo.nlp-tlp.org/data gd.search("A38751") # a report, hole, tenement or site by its id - 03
Data around a point
The commonest question: what is known near here? One call returns drillholes, samples, assay highlights, reports and tenure within a radius.
Now you have: A summary of the ground around the point, and the points themselves as GeoJSON.near = gd.nearby(121.47, -30.75, radius_km=5, metal="Au") for f in gd.nearby_points(121.47, -30.75, radius_km=5, kind="drillhole"): print(f["properties"]) # GeoJSON features, every page - 04
Data under a region
For an area rather than a point: a bounding box gives counts and headline figures; a polygon gives the assays as a file.
Now you have: Figures for the region, the geochemistry inside your polygon as a zip, and a download of the layer under it.gd.region_summary(121.3, -30.9, 121.7, -30.5, metal="Au") ring = [(121.3, -30.9), (121.7, -30.9), (121.7, -30.5), (121.3, -30.5)] gd.geochem_export("company_dh_collars", ring, "collars.zip") # EarthBank-ready zip job = gd.export("drillholes", bbox=(121.3, -30.9, 121.7, -30.5), format="csv") gd.download(job["url"], "drillholes.csv") - 05
Reports and documents
Every WAMEX report has been read page by page. Find the reports that cover a place, then fetch their files.
Now you have: The reports for a place (count is the total, page.next the rest) and download URLs for their PDFs and data files. The WAMEX mirror holds every file DMIRS's two catalogues list, synced weekly, ZIPs kept as ZIPs; a report's listing says what it lacks: complete is false and missing names each file (name, bytes, reason) the mirror does not hold, about 470 files over 325 of 119,000 reports, most of them refused by the department itself (reason unavailable_at_source). Where a report's only PDF is inside one of its ZIPs, that PDF is listed too, with extracted_from naming the ZIP.for r in gd.wamex(121.47, -30.75, nearby=True, radius_km=10): # every page print(r["anumber"], r["year"], r["title"]) files = gd.report_files("A38751") # PDF and data-file links gd.doc_files("A38751") # every file of one document gd.download_reports(["A38751", "A6467"], "wamex/") # every file of many reports into wamex/A<n>/, resumable - 06
Core scans
HyLogger-scanned drillholes from every State's core library, interpreted metre by metre.
Now you have: Hole lists, per-hole mineral logs and tray photos; the same holes open at https://coreflagging.nlp-tlp.org/map?hole=<key>.holes = list(gd.core_holes(state="WA")) hole = gd.core_hole(holes[0]["key"]) # the metre-by-metre mineral log metre = gd.core_metre(hole["key"], 120) - 07
Query the database directly
When the endpoints are not the shape you need, ask the database. Read-only, with a 30-second statement timeout, up to 5,000 rows a page and 30 queries a minute.
Now you have: Rows from any table on the platform, as JSON; give paged queries an ORDER BY so pages do not overlap.gd.tables() gd.describe("mdhdb_raw.gswa_dh_collar") for row in gd.sql("SELECT hole_id, max_depth FROM mdhdb_raw.gswa_dh_collar ORDER BY hole_id", max_rows=5000): print(row["hole_id"], row["max_depth"]) # rows as dicts, every page # with pandas: GeoData.to_dataframe(gd.sql(...)) - 08
Map tiles in your own map
The layers the data map draws are plain tiles; add them to any Mapbox GL or MapLibre map. No token needed.
Now you have: The platform's layers inside your own map.map.addSource("drillholes", { type: "vector", tiles: ["https://platform.nlp-tlp.org/geo/api/tiles/drillholes/{z}/{x}/{y}"], minzoom: 5 }); map.addLayer({ id: "drillholes", type: "circle", source: "drillholes", "source-layer": "drillholes", paint: { "circle-radius": 3 } }); map.addSource("magnetics", { type: "raster", tiles: ["https://platform.nlp-tlp.org/geo/api/raster/magnetics/{z}/{x}/{y}.png"], tileSize: 256 }); // vector layers: drillholes, drillholes_density, company_dh_collars, wamex, minedex, tenements, tenements_dead, // faults, worm_mag, worm_grav, hist, geology, regolith, conservation, native_title, pastoral, heritage // raster layers: magnetics, gravity (zoom 3–8) - 09
Use it in QGIS or ArcGIS
The public layers are also an OGC API – Features service, so a GIS adds them by URL: drillholes and collars (with links to each hole's assays and its CoreFlagging core scan), surface and regional geochemistry, WAMEX report areas, exploration activity, tenements, Minedex, bedrock geology, regolith, faults, worms, conservation reserves, HyLogger scans and reports, and the MineTrace training sites. The rasters come in as XYZ tiles. No token needed; each layer carries its licence, and the faults, bedrock geology, regolith and conservation reserves are non-commercial.
Copy the QGIS, ArcGIS Pro, GDAL Now you have: The platform's layers as live layers in your GIS project, filtered to the area you look at.Service URL https://platform.nlp-tlp.org/ogc All layers https://geo.nlp-tlp.org/gis (every layer, its licence, a sample and its catalog page) QGIS Layer > Add Layer > Add WFS / OGC API - Features Layer New > Name: AgenticGeoMate, URL: https://platform.nlp-tlp.org/ogc, Version: OGC API - Features Connect, pick a collection, Add. Zoom in first: features load for the view, 1,000 a page. ArcGIS Pro Insert > Connections > Server > New OGC API Server, URL: https://platform.nlp-tlp.org/ogc GDAL / ogr ogrinfo OAPIF:https://platform.nlp-tlp.org/ogc ogr2ogr -f GPKG drillholes.gpkg OAPIF:https://platform.nlp-tlp.org/ogc gswa_drillholes -spat 121.3 -30.9 121.7 -30.5 curl curl "https://platform.nlp-tlp.org/ogc/collections?f=json" curl "https://platform.nlp-tlp.org/ogc/collections/hylogger_core_scans/items?f=json&bbox=121.3,-30.9,121.7,-30.5&limit=100" Assays Collar layers carry assays_url: every interval down that hole, as JSON (paged) curl "https://platform.nlp-tlp.org/geo/api/holes/company/<collar id>/assays/" ── Rasters as XYZ tiles ───────────────────────────────────────────────────────────── QGIS Browser panel > XYZ Tiles > right-click > New Connection Name: NTLP magnetics, URL: the template below, Min zoom 3, Max zoom 8, OK Drag the connection onto the map (Web Mercator, EPSG:3857) Layer URL template magnetics https://platform.nlp-tlp.org/geo/api/raster/magnetics/{z}/{x}/{y}.png gravity https://platform.nlp-tlp.org/geo/api/raster/gravity/{z}/{x}/{y}.png ArcGIS Pro Map > Add Data > Data From Path, the same template with {level}/{col}/{row} in place of {z}/{x}/{y} ── MineTrace prospectivity surface (one image per metal, not tiles) ───────────────── Image https://platform.nlp-tlp.org/geo/api/surface/Au.png (Au, Co, Cu, Mn, Ni, REE, Sn, Ta, W; Au_max etc. = upper bound) Corners https://platform.nlp-tlp.org/geo/api/surface/Au/ (west, north, east, south in degrees) GDAL The PNG is in Web Mercator. Georeference it, then open the GeoTIFF in QGIS: gdal_translate -a_srs EPSG:3857 -a_ullr 12567630 -1544447 14363300 -4187029 /vsicurl/https://platform.nlp-tlp.org/geo/api/surface/Au.png au_prospectivity.tif (corners for Au; for the other metals 12569133 -1542900 14361543 -4184346) - 10
Your own data
Upload a CSV or GeoJSON, keep it private or publish it, and share it by link. It joins the map and the assistant like any other layer.
Now you have: A dataset of yours on the platform, with a share link on https://geo.nlp-tlp.org.ds = gd.upload_dataset("kalgoorlie_soils.csv", name="Kalgoorlie soils", visibility="public") # a Read and write token print(f"https://geo.nlp-tlp.org/datasets/{ds['id']}") # the share link - 11
Run a toolbox tool
Geoscience methods from the UWA / Loop3D group and ours, each run on an area you choose: a stereonet of the structural measurements, the geological topology of the map, and more as they come. Quick tools answer at once; long ones run in the background and email you.
Now you have: A tool's results for your area, with a provenance record of the code, version, parameters and inputs that made them. Map layers it makes go to your data library.headers = {"Authorization": "Bearer <token>"} tools = requests.get(f"{P}/api/tools/", headers=headers).json()["tools"] # each with its parameters and paper area = {"point": [117.5, -22.0], "radius_km": 10} # or {"polygon": [[lon, lat], ...]} r = requests.post(f"{P}/tools/api/stereonet/run/", headers=headers, json={"area": area, "params": {"types": ["bedding", "cleavage"], "format": "png"}}).json() for o in r["outputs"]: # a quick tool: outputs inline (text, or base64 for images) print(o["name"], o["format"]) # a long tool answers 202 with a run id: follow it, then fetch each output run = requests.get(f"{P}/tools/api/runs/{r['id']}/", headers=headers).json() link = requests.get(f"{P}/tools/api/runs/{r['id']}/outputs/stats.json", headers=headers).json() # {url} or {dataset_id} - 12
Agents
The same data as MCP tools. Add the address to Claude, Cursor or your own agent; sign in with a platform account when the client asks.
Copy the Claude Desktop / Cursor config Now you have: An agent that can search the documents, query the database and read the drillholes for you.{ "mcpServers": { "geo": { "url": "https://geomate.nlp-tlp.org/mcp" } } }