# PerilScore

> Catastrophe risk scores for US properties. Per-peril (hurricane, wildfire,
> hail, flood, earthquake, tornado) plus a standalone crime peril (excluded
> from the overall property risk), nearest fire station, Protection score,
> and an explicit premium report with property and construction data, COPE,
> indicative replacement-cost (RCV) estimates, permit signals, imagery, and an
> AI-assisted underwriting summary.

## API

- REST full report: GET https://app.perilscore.com/api/v1/score/?address={address} with Bearer API key and available capacity
- Signup: POST https://app.perilscore.com/api/v1/signup/ (JSON body with `email`; magic link → API key)
- MCP (mixed auth, 10 tools, 2025-11-25): POST https://app.perilscore.com/mcp
- Legacy MCP full report (Bearer API key and available capacity): POST https://app.perilscore.com/api/mcp/
- MCP anonymous demo (single tool): POST https://app.perilscore.com/claude/mcp/
- ChatGPT App MCP: POST https://app.perilscore.com/chatgpt/mcp

## Connector tools (mixed-auth /mcp)

- score_address (public, read-only) — free score; never consumes a credit
- build_underwriting_report (score:write) — explicit premium report; requires
  OAuth or a Bearer API key plus an idempotency_key; may consume one credit
- get_score (read) — look up a score by public_id
- list_scores (read) — cursor-paginated history
- get_share_score_url (read) — open the owned human sharing workflow
- get_report_pdf_url (read) — PDF report URL
- get_account_credits (read) — balance + plan info
- export_score_factors (read) — underwriter-ready score factor transparency
- get_acord_url (read) — eligible ACORD PDF URL; defaults to form_type 140 and supports 80

## Auth

OAuth 2.1 + PKCE (Connector Directory): https://app.perilscore.com/.well-known/oauth-authorization-server

Bearer API keys: `Authorization: Bearer ps_api_<32 hex>`. Sign in at
https://app.perilscore.com/developers/api/ to generate a key (or POST https://app.perilscore.com/api/v1/signup/).

`initialize`, `tools/list`, `resources/list`, `resources/read`, and
`score_address` are public. Protected tools trigger OAuth lazy authentication
or accept a PerilScore organization API key. Hosted Claude uses Anthropic-held
OAuth credentials; Claude Code uses Anthropic's trusted CIMD client metadata.

## Tier policy

- Authenticated REST is a full-report operation: a fresh score may consume one
  credit and REST has no caller-supplied idempotency key. Prefer canonical MCP
  when replay-safe creation or explicit charge/reuse metadata matters.
- `score_address`: always free and non-billable; per-peril scores + fire station
  + Protection score only.
- `build_underwriting_report`: protected and explicit; returns normalized
  property/construction data, COPE, indicative RCV, permits where available,
  imagery, AI-assisted findings, and billing metadata. A fresh report may use
  one credit; idempotent replays and recent matches are not charged again.

PerilScore provides decision-support evidence, not a coverage decision. A
qualified insurance professional must verify the data and review any
customer-authored eligibility or underwriting rule before a consumer-impacting
decision is finalized or communicated.

## Discovery

- llms-full.txt: https://app.perilscore.com/llms-full.txt
- OpenAPI spec: https://app.perilscore.com/openapi.json
- MCP manifest: https://app.perilscore.com/.well-known/mcp.json
- Agent skill: https://app.perilscore.com/developers/agent-kit/SKILL.md
- Installable Agent Kit: https://app.perilscore.com/developers/agent-kit/build-with-perilscore.zip
- Public docs: https://app.perilscore.com/developers/

## Support

- Email: info@perilscore.com
- Connector reviewers: connectors@perilscore.com
- Privacy policy: https://app.perilscore.com/privacy/
- Terms: https://app.perilscore.com/terms/