Skip to content
ContextKit

Quick Start

The fastest path: tell your AI agent

Section titled “The fastest path: tell your AI agent”

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

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

The agent will run context setup, introspect your database, and start asking you questions about your data — metrics, ownership, business rules. It reads the generated context/AGENT_INSTRUCTIONS.md and knows exactly what to do.

Or do it yourself

Section titled “Or do it yourself”

  1. Install the CLI

    Terminal window
    npm install -g @runcontext/cli

  2. Run the setup wizard

    Terminal window
    context setup

    The wizard auto-detects your database, introspects tables, scaffolds metadata, enriches to Silver, and generates agent instruction files — all in one interactive flow.

  3. Check your tier

    Terminal window
    context tier

    You should see your model at Silver tier. Gold requires human or agent curation.

  4. Curate to Gold

    Tell your AI agent: “Read context/AGENT_INSTRUCTIONS.md and get this project to Gold.”

    The agent will go back and forth with you, asking about your data as it builds. Or use the visual editor:

    Terminal window
    context dev --studio

  5. Serve to AI agents

    Terminal window
    context serve --stdio

    Your MCP server is live. AI agents now have full semantic context.

From a new project

Section titled “From a new project”
Terminal window
npx create-contextkit my-project
cd my-project
context lint
context tier

From an existing project

Section titled “From an existing project”
Terminal window
npm install -D @runcontext/cli


# Scaffold metadata from a database
context introspect --db duckdb://warehouse.duckdb


# Auto-enrich to Silver
context enrich --target silver --apply --source default


# Check what Gold requires
context tier

What just happened?

Section titled “What just happened?”

The setup wizard performed these steps:

  1. Detected your database (DuckDB, PostgreSQL, MySQL, SQL Server, SQLite, Snowflake, BigQuery, ClickHouse, or Databricks — including databases configured in your MCP server configs)
  2. Introspected every table — generated OSI YAML with columns, types, descriptions
  3. Scaffolded governance files — ownership, security classification, grain statements
  4. Enriched to Silver — added lineage, sample values, glossary links, trust status
  5. Linted the result — flagged anything that still needs attention

Your metadata now lives in YAML files under context/. Edit them directly or let AI agents help curate them toward Gold.