Spendict

Spendict is a verdict layer for teams running AI advertising agents. When your agent can write a hundred ads an hour, the expensive question becomes which three deserve budget — and Spendict answers it with a deterministic run / fixfirst / kill verdict on every ad before a cent is spent, not after. It exposes four tools: assessadcreative scores hook, angle, clarity, audience fit, platform fit, CTA, and compliance, and names the single predicted failure mode; auditcampaignstructure checks budget allocation, audience setup, bid strategy, and measurement against each platform's actual rulebook; analyzecampaignperformance takes live metrics and separates creative fatigue from structural problems; and strategizetargeting returns a structure-validated targeting strategy. Verdicts are deterministic by design: the model proposes dimension scores, but the server recomputes the weighted overall score and derives the verdict through fixed, published gating rules calibrated by performance marketers with over ten years in paid social. MCP support is native — a streamable HTTP connector works in Claude, Cursor, Windsurf, and n8n out of the box — alongside a drop-in Skill for Claude Code, Cursor, Codex, and Gemini, an npm CLI, and one-for-one REST endpoints. Guardrails include quota checks before every inference call, automatic refunds for failed calls, and hashed API keys. Spendict never touches your ad accounts; it only reads what you send and returns a verdict. Pricing starts free with 100 verdicts and no credit card, then $19, $49, and $199 monthly tiers, with overage at roughly a penny a call.