MCP for AI agents
Crime Data API offers an MCP server over Streamable HTTP at https://crimedata.dev/mcp. No API key is required for MCP access. Start with crime_coverage to inspect supported cities, sources, licences and freshness, then request incident statistics. Retain returned provenance and coverage information when presenting results to users or other tools.
Last updated: 2026-11-04
Connect to https://crimedata.dev/mcp using Streamable HTTP. No API key required. Check coverage before interpreting results; reports are not convictions or predictions of personal risk.
crime_coverage: covered cities, sources, licences and freshness.crime_incidents: recent incidents near coordinates or a US address, up to 50 rows.crime_stats: counts by category, month, hour and weekday, plus trends.crime_compare: compare 2–5 coordinate points with the same radius; usepoints: "lat,lon;lat,lon".crime_score: within-city incident density score and grade over the city's last 12 months; outside incident coverage, city-wide FBI fallback when available, with no grade.
Free limits: 60 tool calls per hour per IP, radius at most 1,000 m, no bounding boxes. Incidents, stats and comparisons cover the last 180 days and accept whole dates; incidents return at most 50 rows. Score uses its separate 12-month window. Coordinates or US addresses work for area tools; compare requires coordinates.
Connect: what should you know?
Claude: add the endpoint as a remote connector where your plan supports custom connectors. ChatGPT: add it as a remote MCP app in developer mode where available. Features and access depend on your client and plan.
Cursor: add to .cursor/mcp.json:
{ "mcpServers": { "crime-data": { "url": "https://crimedata.dev/mcp" } } }
For clients requiring an explicit transport:
{ "mcpServers": { "crime-data": { "type": "http", "url": "https://crimedata.dev/mcp" } } }
Example prompts: what should you know?
- Claude: “Check crime_coverage, then summarize reported incident categories within 500 m of Union Square in San Francisco over the last 90 days. Cite the source and freshness.”
- ChatGPT: “Use crime_score at 37.788, -122.4075 with a 500 m radius. Explain the window and why this is not a prediction of personal risk.”
- Cursor: “Use crime_compare for 37.788,-122.4075;37.7599,-122.4148 within 500 m. Explain coverage limitations and show the returned attribution.”
Data comes from each city's open-data portal under the licence shown. It is police-reported incident data, not convictions, and can be revised by the source. Compared with the FBI API · Terms · Refunds · Privacy · DPA · Contact
Operated by Shahar Frish, Israel.
What should I know about this data?
Where does the incident data come from?
Incident data comes from city open-data portals under the licence shown for each city. It is police-reported data, not convictions, and sources can revise it.
Where can I check coverage and freshness?
The city coverage pages and GET /v1/cities identify covered cities, source licences and newest record dates.
Do scores predict personal risk?
No. Scores describe incident density within a city, are not comparable across cities and do not predict personal risk.