Introduction

Turn Your Database Into an AI-Ready Data Product

ContextKit builds a semantic plane between your database and your AI tools — a structured metadata layer that captures what your data means, not just what it’s named. AI agents use it to generate correct SQL on the first try.

The Problem

AI agents see table and column names, but they don’t understand what the data means. They don’t know:

• That revenue should be SUM’d, never AVG’d
• That queries on transactions must filter WHERE status = 'completed'
• That customer_id joins orders to customers as many-to-one
• Exactly which SQL to use for “show me top customers by revenue last quarter”

Every query is a guess. And guesses are wrong often enough to be dangerous.

The Solution

1. context setup          → Browser wizard opens
2. Fill out Context Brief → Name, owner, sensitivity, database
3. Pipeline auto-runs     → Introspect → Scaffold → Enrich → Verify
4. context serve          → MCP server live — agents have full context

Step 1 — Context Brief. Run context setup and a 5-step wizard opens in your browser. Name your data product, set an owner, choose a sensitivity level, and connect your database. Upload any existing documentation (SQL files, data dictionaries, markdown).

Step 2 — Automatic Pipeline. ContextKit introspects your schema, scaffolds Bronze-tier metadata, then auto-enriches toward Silver (trust status, glossary, lineage, sample values). You watch each stage complete in the browser.

Step 3 — Curate to Gold. Open the visual studio (context dev --studio) to add semantic roles, aggregation rules, guardrails, golden queries, and business rules. Or tell your AI agent to curate for you — the generated AGENT_INSTRUCTIONS.md tells it exactly what to do.

Step 4 — Serve via MCP. Run context serve and every AI tool in your stack gets full context. Agents can search metadata, look up business terms, retrieve guardrails, and find golden queries.

Three Ways to Use It

1. Browser wizard (recommended for getting started)

context setup

Opens a 5-step form in your browser. Fill out the Context Brief, connect your database, and the pipeline handles the rest.

2. AI agent runs CLI commands

In Claude Code, Cursor, Windsurf, or any agentic coding platform:

“Install @runcontext/cli and build a semantic layer for my database.”

The agent uses CLI commands and reads AGENT_INSTRUCTIONS.md to build your semantic plane. It goes back and forth with you — asking about metrics, ownership, and business rules.

3. Visual studio for humans

context dev --studio

Browser-based editor for descriptions, semantic roles, golden queries, and business rules. Changes save to disk and the tier scorecard updates live.

As an MCP Server

For agents that need to query an existing semantic plane:

{
  "mcpServers": {
    "contextkit": {
      "command": "npx",
      "args": ["@runcontext/cli", "serve", "--stdio"]
    }
  }
}

Exposes 8 MCP tools: search, explain, validate, tier, golden-queries, guardrails, list-products, get-product.

What’s Included

One install, everything you need:

npm install @runcontext/cli

• Browser setup wizard — 5-step Context Brief form with automatic pipeline
• Visual studio editor — curate metadata in the browser with live tier updates
• 16 CLI commands — introspect, enrich, lint, fix, build, tier, blueprint, serve, and more
• MCP server with 8 tools — serve your semantic plane to AI agents via stdio or HTTP
• 40 lint rules — schema validation, governance, security, data accuracy
• Bronze/Silver/Gold tier system — automated metadata maturity scoring
• AI Blueprints — export Gold-tier metadata as portable OSI YAML
• Static site generator — browsable documentation from your metadata
• 9 database adapters — DuckDB, PostgreSQL, MySQL, SQL Server, SQLite, Snowflake, BigQuery, ClickHouse, Databricks
• Agent instruction generation — setup creates AGENT_INSTRUCTIONS.md so any AI agent knows how to curate

Built on the Open Semantic Interchange (OSI) specification — your metadata is portable, not locked in.