arcgis-mcp

Generic MCP connector for querying ArcGIS REST services - built so it works against RRC's well/pipeline data today, and your own ArcGIS Online account later (once you have a login), without any code changes.

Why generic, not hardcoded to one URL

RRC's own statewide GIS service (gis.rrc.texas.gov) exists but its exact service path wasn't confirmed directly (an automated-access block on that one page stopped me checking it myself). What is confirmed and verified live is a regional mirror of the same TXRRC well/pipeline data, hosted by Houston-Galveston Area Council - see KNOWN_SERVICES in src/arcgis_client.py. That mirror's extent suggests Houston-area coverage rather than confirmed statewide, so treat it as a starting point, and swap in RRC's real statewide URL the moment you confirm it (same tool works against any ArcGIS REST MapServer/FeatureServer URL).

Tools exposed

• list_known_services() - pre-verified service URLs and layer notes.
• get_layer_metadata(service_url, layer_id?, token?) - field names, geometry type, extent for a service or one layer in it. Call this first against any new/unknown service so you know what to filter on.
• query_arcgis_layer(service_url, layer_id, where, out_fields, bbox?, token?)
  • runs a standard Esri REST query, returns GeoJSON. token is optional and only needed for private/ArcGIS Online services later.

Verified so far

get_layer_metadata was checked directly against the live HGAC TXRRC/Wells service and returned real field names (API, WELLID, LONG83, LAT83, etc. on the Surface Wells layer). The actual query_arcgis_layer data call follows the standard, widely-documented Esri REST query spec, but wasn't hit live before handoff - it should work as-is; if a response looks off, send me the error/output and I'll adjust.

Run locally

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
python main.py

Test:

curl -X POST http://localhost:3000/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"query_arcgis_layer","arguments":{"service_url":"https://www.gis.hctx.net/arcgishcpid/rest/services/TXRRC/Wells/MapServer","layer_id":0,"where":"1=1","out_fields":"API,WELLID","bbox":"-96.9,29.0,-96.5,29.4"}}}'

Deploy

Same pattern as before:

git init && git add . && git commit -m "Initial ArcGIS MCP connector"
git remote add origin https://github.com/<you>/arcgis-mcp.git
git push -u origin main

Render dashboard -> New -> Blueprint -> point at the repo (picks up render.yaml) -> deploy. Endpoint will be https://arcgis-mcp.onrender.com/mcp.

Adding your own ArcGIS Online account later

Once you have a login: generate a token (ArcGIS Online -> your account -> generate API key/token, or OAuth if you want it to expire/refresh), then pass it as the token argument on query_arcgis_layer / get_layer_metadata calls against your own hosted feature layers - no redeploy needed, it's just a per-call argument.

Combining with Enverus isopach data

This connector only fetches ArcGIS layers (wells, pipelines, or whatever else you point it at). Combining that with Enverus's isopach output into one map happens at the chat level - ask Claude to pull both and render them together once you're connected to this and have wells/pipelines to overlay.