Quick Start

​

Prerequisites

• Pylar Account: Sign up at pylar.ai
• Database Credentials: Access to at least one supported database or data warehouse
• AI Agent Builder: An agent builder that supports MCP tools (Claude Desktop, Cursor, Windsurf, or custom)

​

Step 1: Connect Your Databases

​

Supported Data Sources

• BigQuery
• Snowflake
• PostgreSQL
• MySQL
• Redshift
• AlloyDB
• Databricks
• And more…

​

Add a Connection

1. Log in to your Pylar dashboard
2. Navigate to Connections → Add Connection
3. Select your database type
4. Enter your connection credentials
5. Test the connection and save

​

Step 2: Create a Project

1. From the dashboard, click Create Project
2. Give your project a name (e.g., “Customer Analytics” or “Error Monitoring”)
3. Optionally add a description
4. Click Create

​

Step 3: Create Your First View

​

Open the SQL IDE

1. In your project, click Create View
2. The SQL IDE opens with your available data sources

​

Select Your Data Source

1. Choose a data source from the dropdown
2. You can see available databases, schemas, and tables
3. Select the database you want to query

​

Write Your Query

• Query Single Tables: Select from individual tables
• Join Multiple Tables: Combine data from multiple tables
• Cross-Database Joins: Join across different databases and warehouses in a single query
• Apply Filters: Use WHERE clauses to filter data
• Aggregate Data: Use GROUP BY, aggregations, and window functions

​

Example: Customer Analytics View

-- Customer Analytics View
-- This view provides aggregated customer data for AI analysis
-- without exposing PII or sensitive financial details

SELECT
  c.customer_id,
  c.region,
  c.signup_date,
  COUNT(o.order_id) as total_orders,
  SUM(o.amount) as total_revenue,
  AVG(o.amount) as avg_order_value,
  MAX(o.order_date) as last_order_date
FROM customers c
LEFT JOIN orders o ON c.customer_id = o.customer_id
WHERE c.is_active = true
GROUP BY c.customer_id, c.region, c.signup_date;

​

Example: Cross-Database Join

-- Join data across multiple databases
SELECT 
  customers.name,
  orders.total,
  payments.status
FROM bigquery.production.customers customers
JOIN snowflake.analytics.orders orders 
  ON customers.id = orders.customer_id
JOIN postgres.payments.payments payments
  ON orders.id = payments.order_id
WHERE orders.date >= '2024-01-01';

​

Save Your View

1. Click Save View when your query is ready
2. Give your view a descriptive name
3. Add a description (this helps when creating MCP tools)
4. Your view is now saved in your project

​

Step 4: Create MCP Tools

​

Method 1: Create with AI (Recommended)

1. Click Create MCP Tool on your view
2. Select Create with AI
3. Write a natural language prompt describing what you want: Example prompts:
   • “Create a tool to fetch error codes, events, and deal information”
   • “Build a tool that returns customer analytics by region”
   • “Make a tool to get order history for a specific customer”
4. Pylar’s AI will:
   • Understand your intent
   • Configure tool parameters
   • Generate tool descriptions
   • Set up the tool schema
5. Review the generated tool configuration
6. Make any adjustments if needed
7. Click Save Tool

​

Method 2: Create Manually

1. Click Create MCP Tool on your view
2. Select Create Manually
3. Configure:
   • Tool Name: Descriptive name for the tool
   • Description: What the tool does
   • Parameters: Input parameters the tool accepts
   • Query Logic: How parameters map to your SQL view
   • Response Format: Structure of returned data
4. Click Save Tool

​

Step 5: Test Your MCP Tools

1. Click on any MCP tool you’ve created
2. Select Test Tool
3. Enter test parameters
4. Click Run Test
5. Review the results:
   • Verify the query executes correctly
   • Check the response format
   • Ensure data is returned as expected
   • Look for any errors

​

Step 6: Publish Your Tools

1. In your project, click Publish
2. Pylar generates:
   • MCP Server Link: A unique URL for your tools
   • Header Token: An authentication token for secure access
3. Copy both the link and the header token

​

Step 7: Connect to Your Agent Builder

​

Add to Agent Builder

1. Open your agent builder (Claude Desktop, Cursor, Windsurf, etc.)
2. Navigate to MCP configuration settings
3. Add a new MCP server:
   • Server URL: Paste your MCP server link
   • Authorization Header: Add your header token (format: Bearer YOUR_TOKEN)
4. Save and restart your agent

​

Verify Connection

1. In your agent, ask it to use your new tool
2. The agent should be able to access your data view
3. Test a few queries to ensure everything works

​

Step 8: Monitor with Evals

​

Access Evals

1. In your Pylar dashboard, click Evals
2. Select your project and view
3. Explore comprehensive analytics

​

What You’ll See

• Interaction Patterns: How agents are using your tools
• Success vs. Errors: Track successful queries and failures
• Error Analysis: Detailed breakdown of errors:
  • What errors occurred
  • When they happened
  • Why they failed
  • How to fix them
• Query Shape: Understand the types of queries agents are making
• Raw Logs: Access full query logs for debugging

​

Use Evals to Improve

1. Adjust View Scope: Modify your SQL views to include or exclude data
2. Refine MCP Tools: Update tool parameters or descriptions
3. Add New Tools: Create additional tools for common use cases
4. Fix Errors: Address issues revealed in the error analysis

​

Iterate and Improve

​

Make Changes Anytime

• SQL Queries: Update your view queries to change data scope
• View Definitions: Modify what data is included
• MCP Tools: Add, remove, or modify tools
• Tool Configurations: Update parameters or descriptions

​

Changes Reflect Automatically

1. Make your changes in the Pylar platform
2. Changes are automatically reflected in all connected agent builders
3. No need to redeploy or reconfigure agents
4. Agents immediately have access to updated views and tools

​

Key Concepts

​

Views Are the Only Access Level

• ✅ Complete control over what data agents can see
• ✅ No risk of agents accessing unauthorized data
• ✅ All queries go through your defined views
• ✅ Full audit trail of all data access

​

Multiple Tools on One View

• Different parameter sets
• Different query filters
• Different use cases
• Different agent interaction patterns

​

Next Steps

• What is Pylar? - Deep dive into Pylar’s capabilities
• Why Pylar? - Understand the benefits
• Explore advanced features like cross-database joins and complex queries
• Build more views and tools for different use cases
• Use Evals to optimize your tools based on real usage