# Offendersearch MCP Server

Search **all 58 US sex-offender registries** — every US state, DC and the
territories — from any MCP client: Claude
Desktop, Claude Code, or any agent runtime that speaks the
[Model Context Protocol](https://modelcontextprotocol.io).

One tool call returns scored, de-duplicated, provenance-tagged records from a
continuously updated national dataset with 100% US coverage. Built on the official
Python [`mcp`](https://pypi.org/project/mcp/) SDK (FastMCP).

## Tools

| Tool | What it does |
|---|---|
| `search_offenders` | Search every registry for one person/location — name, DOB, age, city/state/ZIP, street address, GIS radius, or free text. Returns scored matches with per-registry outcome (`sourceStatus`) and plain-language `warnings`. |
| `batch_search` | Up to 1,000 lookups in one call, row-in/row-out, each through the same engine as a single search. |
| `get_registry_coverage` | The public coverage catalog: every registry code, scope, legal status and health. No API key needed. |

Results carry `matchConfidence`, `matchBasis` (why it matched), DOB verification
state, aliases, addresses, offense detail, and a per-source citation with
`lastCheckedAt` — so an agent can cite exactly which registry said what.

## Auth — an API key, free to get

Create a key at **[offendersearch.app](https://offendersearch.app/sign-up)** — it
works immediately: **25 free searches, no card required.**

- **Local (stdio):** set `OFFENDERSEARCH_API_KEY`.
- **Remote (streamable-HTTP):** each request forwards its own `X-API-Key` header —
  multi-tenant by design.

## Quick start (Claude Desktop, local stdio)

```bash
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
```

`claude_desktop_config.json`:

```json
{
  "mcpServers": {
    "offendersearch": {
      "command": "python",
      "args": ["/path/to/services/mcp-server/server.py"],
      "env": { "OFFENDERSEARCH_API_KEY": "os_live_..." }
    }
  }
}
```

## Remote / self-hosted (streamable-HTTP)

```bash
MCP_TRANSPORT=streamable-http OFFENDERSEARCH_API_KEY="os_live_..." python server.py
# or: uvicorn server:http_app  (see Dockerfile)
```

Client config for a hosted instance (per-request key, no env secret):

```json
{
  "mcpServers": {
    "offendersearch": {
      "url": "https://<your-host>/mcp",
      "headers": { "X-API-Key": "os_live_..." }
    }
  }
}
```

## Example exchange

> **User:** Is a John Doe, born 1980-04-12, on any registry in New Jersey?
>
> **Agent** → `search_offenders(first_name="John", last_name="Doe",
> dob="1980-04-12", state="NJ")`
>
> **Agent:** One candidate matched with a full date-of-birth confirmation
> (`dobVerification: dob_match`, confidence 1.0): John A. Doe, registered in NJ,
> with the official registry record linked. All 58 registries were searched to
> completion (`sourcesIncomplete: 0`).

## Reading results responsibly

- If `counts.sourcesIncomplete > 0`, the records are a **lower bound** — relay
  `warnings[0]` rather than reporting a clean "no match".
- A match is a **candidate**, not an identification — verify DOB and photograph.
- Results are public-record data and **not a consumer report**; do not use them for
  FCRA-covered decisions without appropriate process.

## More

- API docs: https://offendersearch.app/docs
- OpenAPI: https://offendersearch.app/openapi.json
- MCP page: https://offendersearch.app/mcp-server

## Offendersearch resources

- [Sex offender registry search](https://offendersearch.app/) — search every US registry from one dashboard, free tier included
- [Sex offender statistics by state](https://offendersearch.app/sex-offender-statistics) — ranked counts and per-1,000-resident rates for all 56 US jurisdictions
- [Open data downloads](https://offendersearch.app/data) — free aggregate CSVs (by state and by city)
- [Background check API documentation](https://offendersearch.app/docs) — REST reference, normalized record schema, quickstarts
- [State registry guides](https://offendersearch.app/sex-offender-registry) — how each state's registry works and how to search it
- [MCP server guide](https://offendersearch.app/mcp-server) — hosted endpoint, Claude Desktop and ChatGPT connector setup
- [API pricing](https://offendersearch.app/pricing) — metered, $0.15 per call, volume tiers