Manar (منار) — AI Paid Media Ops Platform

Live: https://manar.asktam.net

Stack

• Frontend: Next.js 14 + Tailwind + TypeScript
• Backend: Python FastAPI (manar-mcp container)
• Databases: PostgreSQL (core + tenants) + ClickHouse (metrics/analytics)
• VPS: 76.13.157.7

Architecture

Multi-tenant

• manar_core DB: users, brands, memberships
• manar_tenants DB: per-brand schemas (brand_alluriv, brand_asktam, etc.)
• Each brand schema contains: recommendations, audit_log, ai_memory, brand_context, ad_accounts, ad_credentials

AI Engine

• Recommendations engine: Claude claude-sonnet-4 with execution state memory — prevents conflicting recs
• Execution state: persistent per-campaign strategy memory stored in brand_context
• Anomaly detection: 6 metric types (CPC, CTR, ROAS, impressions, budget_burn, zero_conversions)
• Corrective engine: anomaly → Claude analysis → corrective recs → auto-execute (L2/L3 pilot)
• Budget guard: no budget increases when campaign ROAS < 70% of target
• Conflict prevention: 3 layers — prompt injection, post-generation filter, periodic purge

Transparency

Every recommendation shows its trigger, action type, executability, and source engine.

Containers

Container	Role	Port
manar-mcp	FastAPI REST + MCP server	8092 (API), 8090 (MCP), 8091 (health)
manar-frontend	Next.js UI	3000
manar-core-db	Users/brands registry	5432
manar-tenants-db	Per-brand operational data	5432

Deploy

MCP Backend

docker build -t manar-mcp:latest ./mcp/
docker stop manar-mcp && docker rm manar-mcp
docker run -d --name manar-mcp --network coolify --restart unless-stopped \
  -e CLICKHOUSE_HOST=clickhouse -e CLICKHOUSE_PORT=8123 \
  -e CLICKHOUSE_USER=tam -e "CLICKHOUSE_PASSWORD=Tam@ClickHouse2026!" \
  -e CLICKHOUSE_DB=manar -e MCP_PORT=8090 -e API_PORT=8092 -e HEALTH_PORT=8091 \
  -e "CORE_DB_URL=postgresql://manar:mn_c0re_2026_s3cr3t@manar-core-db:5432/manar_core" \
  -e "TENANTS_DB_URL=postgresql://manar:mn_t3n4nts_2026_s3cr3t@manar-tenants-db:5432/manar_tenants" \
  -e APP_URL=https://manar.asktam.net \
  -e GOOGLE_ADS_CLIENT_ID=... \
  -e GOOGLE_ADS_CLIENT_SECRET=... \
  -e GOOGLE_ADS_DEVELOPER_TOKEN=... \
  --health-cmd "curl -sf http://localhost:8091/health" --health-interval 15s \
  manar-mcp:latest

Frontend

docker build -t manar-frontend:latest ./frontend/
docker compose -f ./frontend/docker-compose.yml up -d --force-recreate

New Brand Onboarding

curl -X POST https://manar.asktam.net/api/alluriv/api/admin/brands/{slug}/bootstrap

Admin Login

• URL: https://manar.asktam.net/login
• Email: loveriver7@gmail.com
• Password: Manar2026!

Key Files

• mcp/api.py — all REST endpoints (~86KB)
• mcp/recommendations_engine.py — Claude prompt + budget guard + conflict filter
• mcp/execution_state.py — per-campaign strategy memory
• mcp/execution_engine.py — 20 Google Ads API actions
• mcp/anomaly_engine.py — ClickHouse anomaly detection
• mcp/corrective_engine.py — anomaly → corrective recommendations
• mcp/scheduler.py — daily 2am automation loop
• frontend/src/app/dashboard/[brand_slug]/ — dashboard pages
• frontend/src/components/RecommendationCard.tsx — rec card with transparency