Charlotte Open City Data

QueensEstate

An MCP server that lets an LLM answer Charlotte residents' everyday questions with official public data from the City of Charlotte Open Data Portal — https://data.charlottenc.gov.

The portal lives at data.charlottenc.gov. It is an ArcGIS Hub site; its ~380 datasets are ArcGIS REST layers run by the City of Charlotte and Mecklenburg County. No API key is required.

What's on the portal

A survey of the catalog (/api/feed/dcat-us/1.1.json) shows these main themes:

Theme	Examples
Public safety	CMPD incidents (870k+ since 2017, updated daily), homicides, traffic stops, fire incident reports, crashes
City services	311 service requests (3.4M), solid-waste collection routes, code enforcement cases, street closures
Planning & zoning	Zoning, rezoning petitions, historic districts, 2040 policy map, floodplains, parcels
Transportation	CATS bus routes and stops, LYNX light rail, park-and-ride, bike lanes, sidewalks, greenways
Places	Libraries, schools, parks, fire/police stations, pharmacies, grocery stores, EV chargers, day care
Government	Capital projects, budget, employee salaries, council and commissioner districts
Demographics	Census block groups/tracts, Quality of Life neighborhood profiles

Tools

The tools are chosen for the questions residents actually ask. Every tool is read-only, and every location parameter accepts either an address ("600 E 4th St, Charlotte") or "latitude,longitude".

My address

Tool	Answers questions like
get_address_profile	"Who is my council member?" "What police division and fire station serve me?" "What's my zoning? Am I in a flood zone or historic district?"
get_trash_and_recycling_schedule	"What day is trash pickup? Is this a recycling week (GREEN/ORANGE)?"
lookup_address	"Is this a valid address? What's its parcel ID and municipality?"

Safety

Tool	Answers questions like
get_crime_near	"Any car break-ins near my apartment this month?"
summarize_crime	"How have robberies in Charlotte changed since 2019?" "Which division has the most burglaries?"
get_traffic_crashes_near	"Is the intersection by my kid's school dangerous?"

City services and neighborhood change

Tool	Answers questions like
get_311_requests_near	"Has anyone already reported this pothole or streetlight?"
get_code_enforcement_cases	"Does the house I'm renting have open housing-code violations?"
get_street_closures	"Are any roads closed on my commute?"
get_capital_projects_near	"What's being built near me, when is it done, and who do I contact?"
get_pending_rezonings	"Is anyone trying to rezone land near my neighborhood?"

Places and transit

Tool	Answers questions like
find_nearby_places	Nearest library, school, park, greenway, pharmacy, grocery store, EV charger, light rail station, bus stop, public Wi-Fi… (18 categories)
get_bus_route	"How often does the 9 run on Sunday evenings?"

Government transparency

Tool	Answers questions like
get_city_budget	"How much does the city budget for Police?"
get_city_salary_stats	"What do Charlotte firefighters earn on average?" (aggregates only, no names)

Anything else in the catalog

Tool	Purpose
search_datasets	Keyword search over the whole portal
describe_dataset	Fields, types, and record count for a layer
query_dataset	SQL-filtered records from any layer, optionally near a location
summarize_dataset	Server-side counts/sums grouped by fields

Setup

Requires Python 3.14 (developed on 3.14.7).

~/.pyenv/versions/3.14.7/bin/python -m venv .venv
.venv/bin/python -m pip install -e . --group dev

Use with Claude Code

claude mcp add queensestate -- /absolute/path/to/queensestate/.venv/bin/queensestate

Use with Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "queensestate": {
      "command": "/absolute/path/to/queensestate/.venv/bin/queensestate"
    }
  }
}

Other transports and debugging

.venv/bin/queensestate --transport http --port 8000        # http://127.0.0.1:8000/mcp
npx @modelcontextprotocol/inspector .venv/bin/queensestate  # interactive inspector

The http transport authenticates its callers with Google OAuth and refuses to start without a public URL, a Google client, and an allow list — see Hosting. To try it locally:

QUEENSESTATE_PUBLIC_URL=http://localhost:8000 \
QUEENSESTATE_GOOGLE_CLIENT_ID=... QUEENSESTATE_GOOGLE_CLIENT_SECRET=... \
QUEENSESTATE_ALLOWED_EMAILS=you@gmail.com \
  .venv/bin/queensestate --transport http --port 8000

Hosting

The hosted server runs on Google Cloud Run at https://queensestate.adamwanninger.com/mcp, and is its own OAuth 2.1 authorization server: it delegates the sign-in to Google purely to learn which account is calling, then checks that address against an allow list and mints its own tokens. Google's tokens are never passed through to the MCP client.

claude mcp add --transport http queensestate https://queensestate.adamwanninger.com/mcp

scripts/deploy-gcp.sh provisions and deploys the whole thing, and re-running it is also how you update. See deploy/GCP.md.

Hosting is only for the HTTP transport. Running the server locally over stdio needs no Google account, no allow list, and none of the cloud dependencies.

Development

.venv/bin/pytest                 # unit + in-memory MCP tests; no network access
.venv/bin/ruff format --check . && .venv/bin/ruff check .
.venv/bin/mypy

Layout:

• src/queensestate/arcgis.py: async ArcGIS REST / Hub client (timeouts, bounded retries, host allowlist)
• src/queensestate/address.py: address matching against the county Master Address layer
• src/queensestate/catalog.py: the curated layers behind the purpose-built tools
• src/queensestate/server.py: the MCP tools

Data notes and limitations

• Geocoding uses Mecklenburg County's Master Address layer, so only addresses inside the county resolve. Coordinates outside the county are rejected.
• Freshness varies by dataset: CMPD incidents, 311 requests, and code enforcement are current within days; the Budget Report dataset currently runs FY2018–FY2023. Tools report what the portal publishes.
• CMPD locations are generalized by the department for privacy.
• 311 data records when requests were received, not when they were resolved.
• Timestamps are ISO 8601 in Charlotte local time with the UTC offset (e.g. 2026-09-06T00:00:00-04:00). The underlying services store UTC.
• Budget amounts include large negative lines (e.g. "00 Non Department", Charlotte Water), so get_city_budget reports both the net total and the sum of positive lines.
• Statistics limits: some city map services ignore resultRecordCount on aggregate queries, so the client enforces row limits itself.
• Live bus arrivals are not on the portal; pair this server with a CATS real-time source. Bus stop IDs from find_nearby_places match CATS stop IDs.
• The generic tools only reach hosts that serve portal datasets (gis.charlottenc.gov, meckgis.mecklenburgcountync.gov, services.arcgis.com, gis.ci.charlotte.nc.us).
• Mecklenburg County's server rejects default Python user agents, so the client sends its own.

Ideas for more tools

• Neighborhood Quality of Life indicators by NPA (income, age, housing, amenities)
• Charlotte Fire incident reports by address block
• CMPD officer traffic-stop statistics
• Sidewalk and bike network gaps near an address
• Tree canopy and land surface temperature by neighborhood

License

MIT - see LICENSE.