# Agent Data Platform

> Machine-readable video generation pricing across providers (fal.ai, Runway,
> Google Gemini API, Kling, Luma, Higgsfield), normalized to USD, with
> provenance (source_url), freshness (verified_at), and confidence on every
> row. REST + MCP. Free tier: 100 calls/month, self-serve.

Base URL: https://agent-data-platform.onrender.com
Auth: X-API-Key header. Mint a free key: POST /v1/keys/signup (no auth, returns key once).
Note: free-tier hosting sleeps; first request may take 30-60s cold start.

## Core endpoints

- POST /v1/estimate — cost estimates cheapest-first.
  Body: {"duration_seconds": 8, "count": 1, "resolution": "1080p", "mode": "text-to-video", "require_audio": false}
  (only duration_seconds required). Each estimate: provider, model, per_job_usd,
  total_usd, confidence (high|medium|low — trust this), data_confidence (source
  reliability), notes, source_url, verified_at, verified_days_ago, tier_notes.
- GET /v1/providers — providers with active model counts.
- GET /v1/pricing?provider=&model= — raw pricing rows with provenance.
- GET /v1/changes?since=ISO8601 — append-only change log (default: last 30 days).
- POST /mcp — MCP Streamable HTTP server (stateless, POST only). Tools:
  estimate_video_cost, compare_providers, get_pricing_changes. Same X-API-Key,
  same metering as REST. Send Accept: application/json, text/event-stream.

## Trust model

- confidence = MIN(data_confidence set at entry, freshness/unit-derived).
- Rows from official docs or in-app verification: high. From third-party
  trackers or support docs: medium. Conflicting/unverified sources: low —
  labeled honestly, never hidden.
- verified_at refreshed weekly even without changes; staleness degrades
  confidence automatically (>14d: medium cap, >30d: low).
- All responses wrapped in {data, meta}; meta.schema_version is date-based and
  bumps on breaking changes.

## Machine-readable spec

- OpenAPI 3.1: /openapi.json
- Human docs: /docs

## Quota / errors

- 401 missing_api_key | invalid_api_key. 429 rate_limited (60 req/min) or
  quota_exceeded (100 calls/month free). Errors: {"error": {"code", "message"}, "meta": {...}}.

Contact: 702executivellc@gmail.com (702 Executive LLC)