The agentic stacks operators already pay for — Cognite Atlas AI, Equinor Echo, custom Claude / GPT agents — need a physical-asset condition data layer to act on. Hullproof is that layer. Served as an MCP tool + JSON-schema-described REST, with confidence + lineage that survives the full agent loop.
Aker BP runs Cognite Atlas AI. Equinor runs Echo. Both are agentic stacks that need structured physical-asset condition data to act on. Hullproof emits that layer as JSON the agents can read directly — no human translation step.
Hullproof exposes itself as a tool agents call. /api/mcp/tools lists tools, /api/mcp/asset-types describes domain coverage, /api/mcp/schema/coating-passport is the JSON Schema. Self-describing argument signatures, idempotent operations, structured error codes agents can reason over.
Every output is a structured CoatingPassport object. No PDF parsing, no OCR step, no human-in-the-loop interpretation. The agent reads the object, the agent acts on the object.
Every finding carries confidence score + source-frame URIs + AI model version. When an agent's downstream action gets challenged (operator dispute, regulator audit, insurance claim), the lineage is intact — the agent's input is defensible all the way back to source frames.
Today Hullproof reads footage with a VLM. Tomorrow Vision-Language-Action (VLA) models will let the same agent see-and-act in one loop. CoatingPassport is the structured intermediate that survives the transition — VLA backends swap in behind the same data contract.
An Atlas AI agent fielding vessel + jacket + monopile + pipeline questions doesn't want one data source per asset type. CoatingPassport's universal schema covers 13 verticals on the same object — one integration, full asset spectrum.
Self-describing endpoints. JSON Schema for all responses. Structured error codes. Idempotent where possible. The MCP tool list and the CoatingPassport schema are public — an agent can discover Hullproof and start calling it without an SDK.
| Method | Path | What it does |
|---|---|---|
| GET | /api/mcp/tools | Tool list — every Hullproof capability exposed as an MCP-callable tool. |
| GET | /api/mcp/asset-types | Domain discovery — which asset types Hullproof currently covers + severity rules per domain. |
| GET | /api/mcp/schema/coating-passport | CoatingPassport JSON Schema — the canonical data contract. |
| GET | /api/passports/{id} | Read a passport. Returns canonical JSON. Falls back to fixtures for demo; reads Firestore for tenant data. |
| POST | /api/passports | Write a passport. Validated against schema, rejects 422 with structured issue list on schema failure. |
| GET | /api/openapi | OpenAPI 3.1+ spec for the full surface. Agent-introspectable. |
A Cognite Atlas AI agent fielding a question about a specific offshore-jacket condition. The agent calls Hullproof, gets a CoatingPassport, and reasons over the structured findings.
# 1. Agent discovers Hullproof tools GET https://hullproof.com/api/mcp/tools # 2. Agent fetches the schema once, caches it GET https://hullproof.com/api/mcp/schema/coating-passport # 3. Agent retrieves a specific passport for the asset in question GET https://hullproof.com/api/passports/demo-offshore-jacket-001 # 4. Agent reasons over the structured findings: # - severity counts per zone # - confidence per finding # - source-frame URIs to cite # - applicable_standards (NORSOK N-005, DNV-RP-C203, ...) # - ai_model_version for audit trail
Pick one asset (vessel, jacket, monopile, pipeline). We emit CoatingPassport from real footage. Your agent stack reads it directly. End-to-end demonstrated on a single asset before any commercial commitment.