generated: '2026-08-18'
method: derived
source: >-
  https://offendersearch.app/mcp-server (the three published MCP tool names and descriptions)
  and https://offendersearch.app/ai-integration (the published LLM tool definition), bound by
  name and semantics to operationIds in openapi/offendersearch-api-openapi.yml.
description: >-
  Binds each agent-facing tool Offendersearch publishes to the REST operation that backs it.
  Confidence is capped at medium for every row: the MCP endpoint does not resolve, so no
  tools/list introspection was possible and no tool inputSchema is published — the mapping is
  by name, description and billing semantics only. Where the mapping IS confirmed by the
  provider's own words (get_registry_coverage described as "the public catalog of every
  registry we cover, with its scope, legal status and health — no API key required", which is
  exactly GET /v1/sources) it is marked high.
surfaces:
  openapi:
    path: openapi/offendersearch-api-openapi.yml
    source: https://offendersearch.app/openapi.json
    operations: 36
    gated: false
  mcp:
    url: https://mcp.offendersearch.app
    gated: true
    gate: DNS does not resolve; private beta, request-access only.
  graphql:
    endpoint: null
  llm_tool_definition:
    url: https://offendersearch.app/ai-integration
    gated: false
crosswalk:
  - tool: search_offenders
    surface: mcp
    category: search
    rest: [syncSearch]
    binding: 1:1
    confidence: medium
    note: >-
      "A name, DOB, location or radius search over nationwide coverage" and "returns the same
      normalized, provenance-tagged records as the REST API" — that is POST /v1/search. Whether
      the tool also fronts the async asyncSearch/getSearch pair for long jobs is not stated.
  - tool: batch_search
    surface: mcp
    category: search
    rest: [batchSearch]
    binding: 1:1
    confidence: medium
    note: >-
      "Up to 1,000 queries in one call, one row in and one row out" matches POST /v1/batch
      exactly, including the 1000-row 413 ceiling and per-row billing.
  - tool: get_registry_coverage
    surface: mcp
    category: metadata
    rest: [listSources]
    binding: 1:1
    confidence: high
    note: >-
      "The public catalog of every registry we cover, with its scope, legal status and health —
      no API key required" is GET /v1/sources, the one anonymous operation in the spec
      (security: []). Free in both surfaces.
  - tool: search_sex_offender_registry
    surface: llm_tool_definition
    category: search
    rest: [syncSearch]
    binding: 1:1
    confidence: high
    note: >-
      The provider states plainly: "The handler calls POST /v1/search". Published input_schema
      exposes only firstName, lastName, dob and state, with lastName required — a deliberate
      subset of the full Query schema.
mcp_only: []
rest_only:
  - operation: asyncSearch
    reason: No published tool submits an asynchronous search or supplies a webhookUrl.
  - operation: getSearch
    reason: No published tool collects an async job by searchId.
  - operation: makeReport
    reason: No tool generates the consolidated verification-report PDF; a binary PDF is an awkward MCP return.
  - operation: makeProof
    reason: Internal proof-document add-on; not exposed as a tool.
  - operation: getProofDoc
    reason: Document fetch by token; not exposed as a tool.
  - operation: getRecord
    reason: No tool fetches a single normalized record by recordId.
  - operation: submitSupportMessage
    reason: Support channel, not a data operation.
  - operation: compatSexoffenderPost
    reason: Legacy offenders.io drop-in shape; no reason to expose it to an agent.
  - operation: compatSexoffenderGet
    reason: Legacy offenders.io drop-in shape; no reason to expose it to an agent.
  - operation: signup
    reason: Account provisioning — session-token surface, out of scope for a data tool.
  - operation: login
    reason: Account provisioning — session-token surface, out of scope for a data tool.
  - operation: getAccount
    reason: Dashboard/session operation.
  - operation: listKeys
    reason: Credential management; deliberately not agent-reachable.
  - operation: createKey
    reason: Credential management; deliberately not agent-reachable.
  - operation: rotateKey
    reason: Credential management; deliberately not agent-reachable.
  - operation: deleteKey
    reason: Credential management; deliberately not agent-reachable.
  - operation: getUsage
    reason: Dashboard/session operation.
  - operation: getTeam
    reason: Dashboard/session operation.
  - operation: inviteMember
    reason: Not available (single-owner accounts).
  - operation: removeMember
    reason: Not available (single-owner accounts).
  - operation: getBilling
    reason: Dashboard/session operation.
  - operation: admin/*
    reason: >-
      Twelve internal ops endpoints behind the separate X-Admin-Key credential
      (/v1/admin/accounts, warm-queries, warm, cache-stats, scraper-health, scraper-runs,
      /admin/ingest, /admin/ingest/rerun-failures, /admin/ingest/status, /admin/freshness,
      /admin/ingest/report). Not part of the public API and not agent-reachable.
coverage:
  tools_published: 4
  tools_mapped: 4
  tools_unmapped: 0
  rest_operations: 36
  rest_operations_covered: 3
  rest_operations_uncovered: 33
  note: >-
    Four published tools map onto three distinct REST operations (search_offenders and
    search_sex_offender_registry both back onto syncSearch). Of the 33 uncovered operations, 12
    are admin-key-only and 12 are account/session/dashboard operations, so the agent-relevant
    gap is 9: the async pair (asyncSearch + getSearch), getRecord, makeReport, makeProof,
    getProofDoc, submitSupportMessage and the two compat operations.