Imported from reason-machines/mcp-skills (skills/ktx-ai-data-context-layer/SKILL.md). Install upstream with npx skills add reason-machines/mcp-skills --skill ktx-ai-data-context-layer. Copyright stays with the author.

ktx AI Data Context Layer Skill

Skill by ara.so — MCP Skills collection.

What is ktx?

ktx is a self-improving context layer that teaches AI agents how to query your data warehouse accurately. It automatically:

• Learns from company knowledge - ingests wiki content, organizes it, removes duplicates, flags contradictions
• Maps the data stack - samples tables, captures metadata, detects joinable columns
• Builds a semantic layer - combines raw tables and metrics through a join graph that resolves chasm and fan traps
• Serves agents at execution - exposes CLI and MCP tools with combined full-text and semantic search

Works with PostgreSQL, Snowflake, BigQuery, ClickHouse, MySQL, SQL Server, and SQLite. Integrates with dbt, MetricFlow, LookML, Looker, Metabase, and Notion.

Installation

Global CLI Installation

npm install -g @kaelio/ktx

Project-Specific Installation

npm install @kaelio/ktx

Quick Setup

ktx setup

This interactive command:

1. Creates or resumes a local ktx project
2. Configures LLM and embedding providers
3. Sets up database connections
4. Configures context sources (dbt, Looker, etc.)
5. Builds initial context
6. Installs agent integration

Project Structure

my-project/
├── ktx.yaml                         # Project configuration
├── semantic-layer/<connection-id>/  # YAML semantic sources
├── wiki/global/                     # Shared business context
├── wiki/user/<user-id>/             # User-scoped notes
├── raw-sources/<connection-id>/     # Ingest artifacts and reports
└── .ktx/                            # Local state and secrets (git-ignored)

Important: Commit ktx.yaml, semantic-layer/, and wiki/. Keep .ktx/ local and git-ignored.

Core Commands

Check Project Status

ktx status

Example output:

ktx project: /home/user/analytics
Project ready: yes
LLM ready: yes (claude-sonnet-4-6)
Embeddings ready: yes (text-embedding-3-small)
Databases configured: yes (warehouse)
Context sources configured: yes (dbt_main)
ktx context built: yes
Agent integration ready: yes (codex:project)

Build Context

# Ingest all configured sources
ktx ingest

# Ingest specific connection
ktx ingest --connection warehouse

# Ingest specific source
ktx ingest --source dbt_main

Search Semantic Layer

# Search for metrics and dimensions
ktx sl "revenue"

# Search with JSON output
ktx sl "customer lifetime value" --json

Search Wiki

# Search wiki pages
ktx wiki "refund policy"

# Search with context
ktx wiki "how do we calculate churn"

MCP Server

# Start MCP server for agent clients
ktx mcp start

# Start with specific project
ktx mcp start --project-dir /path/to/project

# Check MCP status
ktx mcp status

Configuration

ktx.yaml Structure

version: "1"
project:
  name: "analytics"
  description: "Company analytics warehouse"

llm:
  provider: "anthropic"
  model: "claude-sonnet-4-6"
  apiKeyEnvVar: "ANTHROPIC_API_KEY"

embeddings:
  provider: "openai"
  model: "text-embedding-3-small"
  apiKeyEnvVar: "OPENAI_API_KEY"

connections:
  warehouse:
    type: "postgres"
    host: "localhost"
    port: 5432
    database: "analytics"
    user: "readonly_user"
    passwordEnvVar: "DB_PASSWORD"
    ssl: false

sources:
  dbt_main:
    type: "dbt"
    connection: "warehouse"
    manifestPath: "./target/manifest.json"
    catalogPath: "./target/catalog.json"

Environment Variables

Create a .env file in your project root:

# LLM Provider
ANTHROPIC_API_KEY=your_key_here

# Embeddings Provider
OPENAI_API_KEY=your_key_here

# Database Credentials
DB_PASSWORD=your_db_password_here

# Optional: Project directory override
KTX_PROJECT_DIR=/path/to/project

LLM Provider Configuration

Anthropic API

llm:
  provider: "anthropic"
  model: "claude-sonnet-4-6"
  apiKeyEnvVar: "ANTHROPIC_API_KEY"

Google Vertex AI

llm:
  provider: "vertex"
  model: "claude-sonnet-4-6"
  projectId: "my-gcp-project"
  region: "us-central1"
  credentialsEnvVar: "GOOGLE_APPLICATION_CREDENTIALS"

Claude Code Session (Local)

llm:
  provider: "claude-agent-sdk"

Database Connection Examples

PostgreSQL

connections:
  warehouse:
    type: "postgres"
    host: "db.example.com"
    port: 5432
    database: "analytics"
    user: "readonly"
    passwordEnvVar: "POSTGRES_PASSWORD"
    ssl: true

Snowflake

connections:
  snowflake:
    type: "snowflake"
    account: "xy12345.us-east-1"
    warehouse: "COMPUTE_WH"
    database: "ANALYTICS"
    schema: "PUBLIC"
    user: "ktx_user"
    passwordEnvVar: "SNOWFLAKE_PASSWORD"

BigQuery

connections:
  bigquery:
    type: "bigquery"
    projectId: "my-project"
    dataset: "analytics"
    credentialsEnvVar: "GOOGLE_APPLICATION_CREDENTIALS"

Context Source Configuration

dbt

sources:
  dbt_main:
    type: "dbt"
    connection: "warehouse"
    manifestPath: "./target/manifest.json"
    catalogPath: "./target/catalog.json"
    docsPath: "./target/index.html"  # optional

Looker

sources:
  looker:
    type: "looker"
    connection: "warehouse"
    projectPath: "./looker-models"

Metabase

sources:
  metabase:
    type: "metabase"
    connection: "warehouse"
    apiUrl: "https://metabase.example.com"
    apiKeyEnvVar: "METABASE_API_KEY"

Notion

sources:
  notion_wiki:
    type: "notion"
    apiKeyEnvVar: "NOTION_API_KEY"
    databaseIds:
      - "abc123def456"
      - "789ghi012jkl"

Agent Integration

Claude Code

After running ktx setup, the integration is automatic. From your project directory:

What is our total revenue this quarter?

Claude Code will use ktx's semantic layer to query accurately.

Codex

# Install ktx skill in Codex
npx skills add Kaelio/ktx --skill ktx

# Use in any project with ktx.yaml

Cursor / OpenCode

Configure MCP in your editor settings:

{
  "mcpServers": {
    "ktx": {
      "command": "ktx",
      "args": ["mcp", "start", "--project-dir", "/path/to/project"]
    }
  }
}

Semantic Layer Usage

Defining Metrics

Create YAML files in semantic-layer/<connection-id>/:

# semantic-layer/warehouse/revenue.yaml
version: "1"
type: "metric"
name: "total_revenue"
description: "Sum of all order amounts"
sql: "SUM(orders.amount)"
dimensions:
  - "customer_id"
  - "order_date"
filters:
  - "orders.status = 'completed'"
source_table: "orders"

Defining Dimensions

# semantic-layer/warehouse/customer_dimension.yaml
version: "1"
type: "dimension"
name: "customer_segment"
description: "Customer segment based on lifetime value"
sql: |
  CASE
    WHEN total_spent > 10000 THEN 'enterprise'
    WHEN total_spent > 1000 THEN 'mid-market'
    ELSE 'smb'
  END
source_table: "customers"

Join Graph

ktx automatically detects joinable columns. You can override in ktx.yaml:

semantic_layer:
  joins:
    - left_table: "orders"
      right_table: "customers"
      left_column: "customer_id"
      right_column: "id"
      type: "inner"

Wiki Management

Adding Wiki Pages

# Add to global wiki
mkdir -p wiki/global
cat > wiki/global/refund-policy.md <<EOF
# Refund Policy

Customers can request refunds within 30 days.
Full refunds issued if:
- Product not as described
- Technical issues unresolved

Partial refunds (50%) if:
- Customer changed mind
- Alternative solution offered
EOF

User-Scoped Notes

# Add user-specific notes
mkdir -p wiki/user/alice
cat > wiki/user/alice/analysis-notes.md <<EOF
# Q1 Analysis Notes

Revenue spike in March due to new product launch.
Check customer_acquisition_source for details.
EOF

Ingesting Wiki Content

# Rebuild wiki index
ktx ingest

# Search after ingestion
ktx wiki "refund timeline"

Common Patterns

Initial Project Setup

// scripts/setup-ktx.ts
import { execSync } from 'child_process';
import * as fs from 'fs';
import * as path from 'path';

const projectDir = process.cwd();

// Create ktx.yaml
const config = {
  version: "1",
  project: {
    name: path.basename(projectDir),
    description: "Analytics warehouse"
  },
  llm: {
    provider: "anthropic",
    model: "claude-sonnet-4-6",
    apiKeyEnvVar: "ANTHROPIC_API_KEY"
  },
  embeddings: {
    provider: "openai",
    model: "text-embedding-3-small",
    apiKeyEnvVar: "OPENAI_API_KEY"
  },
  connections: {
    warehouse: {
      type: "postgres",
      host: process.env.DB_HOST || "localhost",
      port: parseInt(process.env.DB_PORT || "5432"),
      database: process.env.DB_NAME || "analytics",
      user: process.env.DB_USER || "readonly",
      passwordEnvVar: "DB_PASSWORD"
    }
  }
};

fs.writeFileSync(
  path.join(projectDir, 'ktx.yaml'),
  JSON.stringify(config, null, 2)
);

// Run setup
execSync('ktx setup', { stdio: 'inherit' });

Programmatic Ingestion

// scripts/daily-ingest.ts
import { execSync } from 'child_process';

async function runDailyIngest() {
  console.log('Starting daily ktx ingestion...');
  
  try {
    // Ingest all sources
    execSync('ktx ingest', { 
      stdio: 'inherit',
      env: { ...process.env, KTX_PROJECT_DIR: '/path/to/project' }
    });
    
    console.log('Ingestion complete');
  } catch (error) {
    console.error('Ingestion failed:', error);
    process.exit(1);
  }
}

runDailyIngest();

Custom Metric Definition Workflow

// scripts/add-metric.ts
import * as fs from 'fs';
import * as path from 'path';
import * as yaml from 'yaml';

interface MetricDefinition {
  version: string;
  type: 'metric';
  name: string;
  description: string;
  sql: string;
  dimensions?: string[];
  filters?: string[];
  source_table: string;
}

function addMetric(
  connectionId: string,
  metric: Omit<MetricDefinition, 'version' | 'type'>
) {
  const metricDef: MetricDefinition = {
    version: "1",
    type: "metric",
    ...metric
  };
  
  const dir = path.join(
    process.cwd(),
    'semantic-layer',
    connectionId
  );
  
  fs.mkdirSync(dir, { recursive: true });
  
  const filename = `${metric.name}.yaml`;
  const filepath = path.join(dir, filename);
  
  fs.writeFileSync(
    filepath,
    yaml.stringify(metricDef)
  );
  
  console.log(`Created metric: ${filepath}`);
}

// Usage
addMetric('warehouse', {
  name: 'daily_active_users',
  description: 'Count of unique users per day',
  sql: 'COUNT(DISTINCT user_id)',
  dimensions: ['event_date'],
  filters: ['event_type = \'login\''],
  source_table: 'user_events'
});

Searching Programmatically

// scripts/search-context.ts
import { execSync } from 'child_process';

function searchSemanticLayer(query: string): any {
  const result = execSync(`ktx sl "${query}" --json`, {
    encoding: 'utf-8',
    env: { ...process.env, KTX_PROJECT_DIR: '/path/to/project' }
  });
  
  return JSON.parse(result);
}

function searchWiki(query: string): any {
  const result = execSync(`ktx wiki "${query}" --json`, {
    encoding: 'utf-8',
    env: { ...process.env, KTX_PROJECT_DIR: '/path/to/project' }
  });
  
  return JSON.parse(result);
}

// Usage
const revenueMetrics = searchSemanticLayer('revenue');
console.log('Revenue metrics:', revenueMetrics);

const policies = searchWiki('refund policy');
console.log('Policies:', policies);

MCP Integration Details

Available MCP Tools

When ktx MCP server is running, agents have access to:

1. search_semantic_layer - Search metrics, dimensions, and tables
2. search_wiki - Search wiki pages and documentation
3. get_metric_definition - Get full metric SQL and metadata
4. list_connections - List available database connections
5. get_table_schema - Get table column details
6. get_join_paths - Find join paths between tables

Example MCP Usage from Agent

// Agent uses MCP to find revenue metric
const result = await useMcpTool('ktx', 'search_semantic_layer', {
  query: 'total revenue by customer segment'
});

// Get full metric definition
const metricDef = await useMcpTool('ktx', 'get_metric_definition', {
  metric_name: 'total_revenue'
});

// Find join path
const joinPath = await useMcpTool('ktx', 'get_join_paths', {
  from_table: 'orders',
  to_table: 'customers'
});

Troubleshooting

ktx status shows "Project ready: no"

# Check ktx.yaml exists
ls -la ktx.yaml

# If missing, run setup
ktx setup

# Verify project directory
echo $KTX_PROJECT_DIR

LLM provider not configured

# Check environment variables
env | grep ANTHROPIC_API_KEY
env | grep OPENAI_API_KEY

# Add to .env file
echo "ANTHROPIC_API_KEY=your_key" >> .env
echo "OPENAI_API_KEY=your_key" >> .env

# Re-run setup
ktx setup

Database connection fails

# Test connection manually
psql -h localhost -U readonly -d analytics

# Check ktx.yaml credentials
cat ktx.yaml | grep -A 10 connections

# Verify environment variable
env | grep DB_PASSWORD

# Try with explicit project dir
ktx ingest --project-dir /path/to/project

MCP server won't start

# Check if already running
ktx mcp status

# Stop existing server
pkill -f "ktx mcp"

# Start with debug output
ktx mcp start --verbose

# Check MCP logs
tail -f ~/.ktx/logs/mcp.log

Context ingestion fails

# Run with verbose output
ktx ingest --verbose

# Check specific source
ktx ingest --source dbt_main --verbose

# Verify source paths
ls -la target/manifest.json
ls -la target/catalog.json

# Check connection separately
ktx test-connection warehouse

Search returns no results

# Rebuild context
ktx ingest

# Check if files exist
ls -la semantic-layer/
ls -la wiki/

# Try broader search
ktx sl "revenue" --verbose
ktx wiki "policy" --verbose

Permission errors

# Check file permissions
ls -la ktx.yaml
ls -la .ktx/

# Fix ownership
chown -R $USER:$USER .ktx/

# Re-initialize
rm -rf .ktx/
ktx setup

Agent can't find ktx

# Ensure MCP server is running
ktx mcp status

# If not, start it
ktx mcp start --project-dir $(pwd)

# Restart agent client (Claude Code, Cursor, etc.)

# Verify MCP configuration in agent settings
cat ~/.config/claude-code/mcp.json

Telemetry Opt-Out

# Disable telemetry
export KTX_TELEMETRY_DISABLED=1

# Or add to .env
echo "KTX_TELEMETRY_DISABLED=1" >> .env

# Verify
ktx status

Advanced Usage

Custom Join Logic

# ktx.yaml
semantic_layer:
  joins:
    - left_table: "orders"
      right_table: "customers"
      left_column: "customer_id"
      right_column: "id"
      type: "left"
    - left_table: "orders"
      right_table: "products"
      left_column: "product_id"
      right_column: "id"
      type: "inner"
      # Prevent fan-out
      cardinality: "many_to_one"

Multi-Database Setup

connections:
  warehouse:
    type: "postgres"
    host: "warehouse.example.com"
    database: "analytics"
    # ... credentials
  
  production:
    type: "postgres"
    host: "prod.example.com"
    database: "app_db"
    # ... credentials
  
sources:
  dbt_warehouse:
    type: "dbt"
    connection: "warehouse"
    manifestPath: "./warehouse/target/manifest.json"
  
  dbt_production:
    type: "dbt"
    connection: "production"
    manifestPath: "./production/target/manifest.json"

CI/CD Integration

# .github/workflows/ktx.yml
name: ktx Context Build

on:
  push:
    branches: [main]
  schedule:
    - cron: '0 2 * * *'  # Daily at 2 AM

jobs:
  build-context:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      
      - uses: actions/setup-node@v4
        with:
          node-version: '20'
      
      - name: Install ktx
        run: npm install -g @kaelio/ktx
      
      - name: Build context
        env:
          ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
          OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
          DB_PASSWORD: ${{ secrets.DB_PASSWORD }}
        run: |
          ktx ingest
      
      - name: Commit updated context
        run: |
          git config user.name "ktx Bot"
          git config user.email "bot@example.com"
          git add semantic-layer/ wiki/
          git diff --quiet || git commit -m "Update ktx context"
          git push

Best Practices

1. Version Control: Commit ktx.yaml, semantic-layer/, and wiki/ but git-ignore .ktx/
2. Read-Only Access: Configure database connections with read-only users
3. Regular Ingestion: Run ktx ingest daily or on data model changes
4. Metric Naming: Use clear, consistent names (e.g., total_revenue, not rev)
5. Documentation: Document business logic in wiki pages, not just metrics
6. Environment Variables: Never commit secrets; use env vars for all credentials
7. Testing: Test new metrics and joins before committing to version control
8. MCP Management: Keep MCP server running for active agent sessions