DataMCP — AI Data Engineer

Connect your Postgres database. Get a production dbt project in under 10 minutes.

DataMCP is an MCP server that acts as an AI data engineer inside Claude Code, Cursor, and Codex. It connects to your Postgres database, reads and maps the schema, infers the business domain, and generates a complete, production-quality dbt project — automatically.

Full documentation: build and view locally with pip install -e '.[docs]' && mkdocs serve, then open http://127.0.0.1:8000. Covers installation, a full pipeline walkthrough with real output, every tool, AI provider setup, security, and troubleshooting.

Quickstart

pip install datamcp-ai

datamcp init
datamcp configure   # stores credentials in OS keychain, never on disk

# Open Claude Code and type:
# "Set up analytics for my project"

That's it. DataMCP calls the 10 core pipeline tools in sequence (of 30 MCP tools total — see below) and produces:

• Staging models (one per source table, DuckDB-validated)
• Fact and dimension mart models
• 5–11 domain-specific KPI definitions
• A dlt ingestion pipeline
• A Dagster project with a daily schedule
• 30+ data quality tests — all passing

Total time: under 10 minutes.

Installation

pip install datamcp-ai

Requires Python 3.11+. Bring your own Gemini API key (OpenAI optional). The PyPI distribution is named datamcp-ai (the plain datamcp name was already taken by an unrelated package) — the installed CLI command and Python import are still datamcp.

Configuration

datamcp configure
# Prompts for: DB host, port, database, username, password, Gemini key
# (OpenAI key optional — only needed if DATAMCP_AI_PROVIDER=openai)
# All credentials stored in OS keychain — never on disk

datamcp install --editor claude-code  # or cursor, codex

MCP Tools (30)

Core pipeline (steps 1–10 of the standard flow)

Explore & query

Schema analysis & scaffolding

Orchestration, BI & streaming

Security

• Database credentials: OS keychain only (Mac Keychain / Linux Secret Service / Windows Credential Manager)
• AI inference: schema metadata only — table names, column names, types, FK relationships. Never actual data values, credentials, or PII.
• Knowledge base: schema metadata and model definitions only. No row data.

See SECURITY.md and DATA_PRIVACY.md.

What gets generated

datamcp_output/
├── dbt/
│   └── my_project/
│       ├── dbt_project.yml
│       ├── profiles.yml          # Postgres target
│       ├── profiles_duckdb.yml   # DuckDB local dev target
│       └── models/
│           ├── staging/
│           │   ├── sources.yml
│           │   ├── stg_users.sql
│           │   ├── stg_orders.sql
│           │   └── ...
│           └── marts/
│               ├── fct_orders.sql
│               ├── dim_customers.sql
│               └── _metrics.yml
├── dlt/
│   ├── pipeline.py
│   └── sources/my_project.py
└── dagster_datamcp/
    ├── assets/dbt_assets.py
    ├── schedules/daily_schedule.py
    └── definitions.py

Running locally

# Run dbt against DuckDB (no Postgres needed)
cd datamcp_output/dbt/my_project
dbt run --target duckdb
dbt test --target duckdb

# Or run the full Dagster pipeline
cd datamcp_output/dagster_datamcp
dagster dev

Contributing

See CONTRIBUTING.md.

License

Apache 2.0