Adform Media Plans

Adform media plans and forecasting

List, inspect, forecast, and get AI recommendations for media plans. Read-only.

Connection & tooling

Runs on the Adform GraphQL MCP. Use graphql_execute to run queries. Use graphql_search to look up field names; use graphql_introspect for complex nested types. Keep calls sequential (~1–2s apart).

Forecast KPI queries can take 5–15 seconds on complex inputs — allow time for the response.

---

1. List media plans

{
  mediaPlans(
    advertiserIds: ["71883"]
    pagination: { offset: 0, limit: 20 }
  ) {
    mediaPlans { id name advertiserId comment schedule { startDate endDate timeZoneId } budget { currency amount } }
    totalCount
  }
}

2. Get media plan — full detail

{
  mediaPlan(id: "12345") {
    id name advertiserId comment
    schedule { startDate endDate timeZoneId }
    budget { currency amount }
  }
}

Use graphql_introspect on MediaPlan to discover available sub-fields for goals, targeting, schedules, and line items.

3. Forecast KPIs — projected CTR, impressions, cookies, spend

Build a ForecastingKpisMediaPlanInput from an existing plan or from scratch. If referencing an existing plan, fetch it first and apply only the overrides the trader asked for.

> CRITICAL: Use the named-variable pattern only. The inline pattern (inputs baked into the > query body with unquoted enum literals) validates against the schema but returns a 500 server > error at execution due to MCP layer serialisation. Always pass query and variables as > separate arguments to graphql_execute.

Query string (pass as query):

query ForecastScenario($mediaPlan: ForecastingKpisMediaPlanInput!, $currencyCode: CurrencyCode) {
  mediaPlanForecastingKpis(mediaPlan: $mediaPlan, currencyCode: $currencyCode) {
    ctr { kpi impressions spend maxPrice }
    forecast { impressions cookies }
  }
}

Variables (pass as variables, JSON):

{
  "currencyCode": "EUR",
  "mediaPlan": {
    "advertiserId": "71883",
    "timeZoneId": "Europe/Berlin",
    "name": "Forecast scenario",
    "schedules": [
      { "startDateTime": "2026-04-01T00:00:00", "endDateTime": "2026-04-14T23:59:59" }
    ],
    "budget": {
      "currency": "EUR",
      "amount": 10000,
      "frequencyCappingRules": []
    },
    "goals": {
      "kpis": [{ "performance": { "ctr": { "rate": 0.005 } } }]
    },
    "targeting": {
      "idFusionEnabled": true,
      "inventory": { "channels": ["display", "native"] },
      "targetingRuleGroups": [{
        "locations": [{
          "targetingMode": "include",
          "locations": { "countryIds": ["276"] }
        }]
      }]
    }
  }
}

All enum values are quoted strings in the variables JSON: "EUR", "display", "native", "include" etc. currencyCode is at the top level of variables, not inside mediaPlan. frequencyCappingRules must always be an explicit empty array []. Omit inventorySources unless the trader explicitly names specific sources.

4. Get AI optimisation recommendations

The mediaPlan input requires advertiserId and goals at minimum. Fetch the plan first, then pass its parameters. The filter field is recommendationTypes.

{
  mediaPlanRecommendations(
    mediaPlan: {
      advertiserId: "71883"
      goals: { kpis: [{ performance: { ctr: { rate: 0.005 } } }] }
      schedule: { startDate: "2026-07-01", endDate: "2026-07-31", timeZoneId: "Europe/Berlin" }
    }
    filter: { recommendationTypes: [] }
  ) {
    recommendationGroup
    recommendationType
  }
}

Use graphql_introspect on MediaPlanRecommendationsMediaPlanInput and MediaPlanRecommendationsFilterInput to discover available recommendation types.

---

Common workflows

• Plan review: list plans → get plan detail → forecast KPIs to project outcomes →

get recommendations for optimisation tips

• Budget scenario: fetch plan → modify budget in the forecast input → compare results
• Geo scenario: change targetingRuleGroups.locations in the forecast input → see how

reach changes

Presenting

Show plan summary with name, status, and key settings. For forecasts, present projected impressions, unique cookies, CTR, and spend clearly. For recommendations, list type and description. State that forecasts are estimates.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: adform
• Source: adform/agentic-skills
• License: MIT
• Homepage: https://site.adform.com

Install and usage instructions live in the source repository linked above.