unified-ad-intelligence-mcp 0.1.1

Unified Ad Intelligence MCP

Unified Ad Intelligence MCP

Core Problem

Each ad platform claims credit independently, causing duplicate conversion credit and wasted budget.

What This Server Provides

This folder contains a production-minded MCP server scaffold with typed JSON tool responses, connector health metadata, OAuth-oriented configuration, rate-limit guards, stable error envelopes, and deterministic demo data until live vendor integrations are attached.

This server has received a production pass. It now includes local attribution intelligence models for cross-platform spend, platform-reported revenue, deduplicated revenue truth, Shopify revenue truth, marginal ROAS, budget allocation, cohort ROAS, and campaign anomaly detection.

Connectors

• Google Ads API via OAuth scope adwords.readonly; env prefix GOOGLE_ADS
• Meta Marketing API via OAuth scope ads_read; env prefix META_MARKETING
• TikTok Ads API via OAuth scope advertiser.read; env prefix TIKTOK_ADS
• LinkedIn Campaign Manager API via OAuth scope r_ads; env prefix LINKEDIN_ADS
• Klaviyo API via OAuth scope campaigns:read; env prefix KLAVIYO
• Shopify Admin API via OAuth scope read_orders; env prefix SHOPIFY
• Pinterest Ads API via OAuth scope ads:read; env prefix PINTEREST_ADS

MCP Tools

• get_cross_platform_summary - Summarize spend, attributed revenue, true revenue, and duplication risk across platforms.
• get_true_roas - Compare platform-reported ROAS against deduplicated revenue truth.
• get_budget_recommendation - Recommend budget shifts using marginal ROAS and confidence.
• get_new_vs_returning_roas - Split acquisition quality by new and returning customers.
• detect_campaign_anomalies - Detect spend spikes, CPA jumps, conversion drops, and tracking outages.

Current Local Capabilities

• Compares platform-reported revenue against deduplicated true revenue.
• Calculates reported ROAS, true ROAS, overclaim amount, and overclaim rate.
• Recommends budget allocation by profit, acquisition, or scale goals.
• Splits ROAS by new versus returning customer revenue.
• Detects CPA jumps, spend spikes, conversion-rate drops, and tracking anomalies.
• Returns stable errors for unknown platforms, invalid goals, bad budgets, and invalid lookback windows.

Test

python -m pytest .\tests -q -p no:cacheprovider

Partial Platform Support

Customers only need to connect the ad platforms they actually use. Missing Google Ads, Meta, TikTok, LinkedIn, Klaviyo, Shopify, or Pinterest credentials do not prevent the server from starting or running local intelligence tools.

Use get_live_connector_status to see configured connectors. Use test_meta_marketing_connection and test_google_ads_connection for read-only live smoke checks when those platforms are connected.

Running Locally

powershell python -m venv .venv .\.venv\Scripts\pip install -e . .\.venv\Scripts\python -m unified_ad_intelligence_mcp.server

Claude Desktop Config

json { "mcpServers": { "unified-ad-intelligence": { "command": "python", "args": ["-m", "unified_ad_intelligence_mcp.server"], "cwd": "D:\CUSTOMS\EARNALL\MCP Servers\01-unified-ad-intelligence" } } }

Production Checklist

• Create OAuth 2.1 apps for each connector and set *_CLIENT_ID plus *_CLIENT_SECRET.
• Store refresh tokens in a secrets manager, never in repo files.
• Replace deterministic demo rows in server.py with API adapter calls.
• Persist raw snapshots and normalized warehouse tables for trend analysis.
• Add integration tests with recorded fixtures for every connector.
• Validate every tool in MCP Inspector before publishing.
• Deploy with a /health route through FastMCP streamable HTTP on Railway or Render.