🔄 Claude Code for Data Engineers: MCP Driven Data Modeling with dbt & Miro & PostgreSQL

Go from a business PRD on a Miro board to documented dbt models in one Claude Code session.

All context in this post: Claude Code for Data Engineer: MCP Driven Data Modeling with dbt & Miro & PostgreSQL

🧩 Stack

• Miro MCP — Read business requirements, push ERDs for approval
• MCP Data Toolbox — Database exploration (schema discovery, sample data, queries)
• dbt Agent Skills — Model generation with dbt best practices
• Custom Skill: prd-to-dbt — Translates business language → technical implementation plan

🌿 Branches

• main — Starting point. Base dbt project with campaigns, sessions, and conversions modeled. The PRD lives on Miro but nothing from it has been implemented yet. Run the workflow here.
• PR #1 — Full implementation: 2 intermediate + 2 new mart models generated from the PRD.

⚡ Quick Start

# 1. Start PostgreSQL and seed synthetic marketing data
docker compose up -d postgresql
docker compose up seed

# 2. Run dbt
docker compose run --rm dbt dbt compile --profiles-dir . --project-dir .
docker compose run --rm dbt dbt test --profiles-dir . --project-dir .

Postgres MCP (Google GenAI Toolbox)

The .mcp.json is pre-configured for the Docker PostgreSQL instance. Download the toolbox binary:

# macOS ARM64 (Apple Silicon)
curl -O https://storage.googleapis.com/genai-toolbox/v0.7.0/darwin/arm64/toolbox
chmod +x toolbox && mkdir -p bin && mv toolbox bin/

# macOS Intel
curl -O https://storage.googleapis.com/genai-toolbox/v0.7.0/darwin/amd64/toolbox
chmod +x toolbox && mkdir -p bin && mv toolbox bin/

# Linux x86_64
curl -O https://storage.googleapis.com/genai-toolbox/v0.7.0/linux/amd64/toolbox
chmod +x toolbox && mkdir -p bin && mv toolbox bin/

Miro MCP & dbt Agent Skills

Miro MCP is pre-configured in .mcp.json. On first use, Claude will prompt you to authenticate via browser.

# dbt Agent Skills
/plugin marketplace add dbt-labs/dbt-agent-skills
/plugin install dbt@dbt-agent-marketplace

🔄 The Workflow

1. Read PRD + explore database — Miro MCP extracts requirements, Data Toolbox explores raw tables
2. Translate to plan — Custom Skill maps business concepts to models, columns, joins
3. Propose schema on Miro — ERD on the board for stakeholder approval
4. Implement & validate — dbt Agent Skills generate models with schema docs and tests

🗺️ Sitemap

workflow/                          # Full demo walkthrough
  DEMO.md                          #   Master guide — start here
  setup/                           #   Stage 1: Connect tools
    SETUP.md                       #     Master setup guide
    miro/                          #     Miro MCP connection + screenshots
    dbt/                           #     dbt Agent Skills install + screenshots
  design/                          #   Stage 2: Read PRD + propose ERD
    DESIGN.md                      #     Schema design walkthrough
  modeling/                        #   Stage 3: Implement dbt models
    MODELING.md                    #     dbt implementation walkthrough

.claude/skills/prd-to-dbt/        # Custom translation Skill
seed/                              # S3 → PostgreSQL data loader
dbt/                               # marketing_analytics dbt project
  models/
    staging/                       #   4 staging models
    intermediate/                  #   2 intermediate models (int_customer_lifetime_value, int_campaign_funnel)
    marts/                         #   4 marts (campaign_performance, daily_summary, customer_lifetime_value, campaign_funnel_analysis)

The business PRD lives on this Miro board — Claude reads it directly via Miro MCP during the workflow.

Follow the demo: workflow/DEMO.md

📩 Subscribe to The Pipe & The Line