The MCP server for marketing judgment

MCP (Model Context Protocol) is the open standard that lets AI agents call external tools. A marketing MCP server exposes marketing capabilities as tools an agent can invoke mid-task: while your agent writes ad copy, plans a campaign, or reviews performance data, it can call the server and get an expert answer back — no copy-pasting into another app.

Most marketing AI tools generate moreassets. Spendict does the opposite job: it judges them. Agents are good at producing twenty ad variants; they are unreliable at grading their own homework. Spendict is the independent grader — a spend gate between “the agent made an ad” and “the ad gets a budget.”

• ✓assess_ad_creative — judges an ad before a cent is spent and returns run / fix_first / kill, with scores across seven dimensions and the one fix that matters most.
• ✓audit_campaign_structure— checks a campaign build against the platform’s rulebook before budget flows through it.
• ✓analyze_campaign_performance — reads live metrics against benchmarks and names the single bottleneck.
• ✓strategize_targeting — designs one opinionated, executable targeting strategy from a product, budget, and objective.

The verdicts are deterministic: the server recomputes every score and recommendation from fixed gates, so the same ad gets the same verdict — an agent can safely automate against it. Full input/output reference in the docs.

In Claude: Settings → Connectors → Add custom connector, then paste:

https://www.spendict.com/api/mcp

In Claude Code:

claude mcp add --transport http spendict https://www.spendict.com/api/mcp

In Cursor or Windsurf, add { "spendict": { "url": "https://www.spendict.com/api/mcp" } } to your MCP config. Authorization is OAuth — no API key to paste. The same tools are also available over plain REST and a CLI (npm i -g spendict) if your stack doesn’t speak MCP yet.

The cost of a bad ad isn’t the generation — it’s the spend. An agent that launches a weak hook burns real budget for days before the metrics say so. A verdict before launch costs cents and catches the failure while it’s still free to fix. That’s the whole product: seven scored dimensions (hook, angle, clarity, audience resonance, platform fit, CTA, compliance), hard gates a model can’t sweet-talk its way past, and a recommendation your agent can branch on.