[FEATURE] Add MCP (Model Context Protocol) Support

[armalite]

● [FEATURE] Add MCP (Model Context Protocol) Support - Composite Data Quality Hub

Overview

Add Model Context Protocol (MCP) server support to dbt-ai, transforming it into a hostable composite data quality hub that integrates with AI coding agents and aggregates insights from multiple data
tools.

Core MCP Tools

Expose dbt-ai's existing functionality as MCP tools:

1. analyze_dbt_model(model_name): Analyze a single dbt model for quality issues
2. check_metadata_coverage(): Check which models are missing metadata
3. get_project_lineage(): Get model dependencies and relationships
4. assess_data_product_quality(model_name): Generate comprehensive quality assessment

Deployment Architecture

Local Development

Existing CLI (unchanged)

dbt-ai -f ./project --output json

Local MCP server (stdio)

dbt-ai --mcp-server -f ./project

Hostable MCP server (network accessible)

dbt-ai serve -f ./project --mcp-host 0.0.0.0 --mcp-port 8080

Production Deployment

Docker container

docker run -p 8080:8080 dbt-ai:latest serve --mcp-host 0.0.0.0

Kubernetes with Helm

helm install dbt-ai ./charts/dbt-ai/

Composite MCP Architecture

Beyond basic dbt analysis, integrate with other MCP servers for enhanced context:

Phase 2: Git Integration

• Connect to official Git MCP server
• Enhanced tool: analyze_dbt_model_with_git_context(model_name)
• Provides: commit history, blame info, recent changes for models

Phase 3: Future Data Tool Integrations

• Monte Carlo MCP: Data quality metrics and alerts
• DataHub MCP: Data catalog and lineage information
• Snowflake MCP: Query performance and usage analytics

Value Proposition: Single MCP endpoint that aggregates insights from your entire data stack

Implementation Approach

Technical Details

• Framework: FastMCP 2.0+ (Prefect's implementation)
• Architecture: Composite server (client + server capabilities)
• Backwards Compatibility: Existing CLI interface completely preserved
• Response Format: Structured JSON for all tools
• Transport: Support both stdio and HTTP/network modes

Deployment Strategy

• Containerization: Docker with multi-stage build
• Kubernetes: Production-ready deployment manifests
• Helm Chart: Configurable K8s deployment with secrets management
• Scaling: Stateless design for horizontal scaling

Implementation Phases

• Phase 1: Basic MCP server (4 core tools) - COMPLETE ✅
• Phase 2: Hostable server + containerization
• Phase 3: Git MCP integration (composite architecture)
• Phase 4: Container orchestration (Docker + K8s + Helm)
• Phase 5: Additional MCP server integrations (Monte Carlo, DataHub, etc.)

Expected Benefits

For AI Agents

• Instant compatibility with Claude Code, Cursor, and other MCP-enabled agents
• Comprehensive data context from single endpoint (dbt + git + quality + performance)
• Production-ready deployment for team/enterprise use

for Data Teams

• Programmatic access for CI/CD pipelines and monitoring
• Centralized data quality hub accessible from anywhere
• Agentic workflow enablement for data platform automation

for Platform Teams

• Kubernetes-native deployment with standard DevOps practices
• Horizontal scaling for enterprise workloads
• Integration point for broader data platform observability

Competitive Positioning

• First composite dbt MCP server that aggregates multiple data tools
• Production hosting capability (not just local development)