looker-analyst

Looker BI Analyst Toolkit

Your expert guide for data analysis and visualization using Looker, with integrated MCP tooling for programmatic query building and dashboard creation.

Table of Contents

1. Quick Start
2. Core Workflows
3. Looker MCP Integration
4. When to Use This Skill
5. Reference Materials

---

Quick Start

Using Looker MCP Tools

The Looker MCP provides programmatic access to your Looker instance. Start every workflow by discovering available models and explores:

1. Call mcp__looker-toolbox__get_models to list available models
2. Call mcp__looker-toolbox__get_explores with a model to see available explores
3. Call mcp__looker-toolbox__get_dimensions and get_measures to discover fields
4. Build queries using the query tool

Basic Query Pattern

// Standard query structure
{
  "model": "model_name",           // From get_models
  "explore": "explore_name",       // From get_explores
  "fields": [                      // From get_dimensions/get_measures
    "dimension.field_name",
    "measure.field_name"
  ],
  "filters": {                     // Optional
    "dimension.field": "value"
  },
  "sorts": ["field desc 0"],       // Optional
  "limit": 500                     // Optional, default 500
}

---

Core Workflows

1. Exploratory Data Analysis

Goal: Understand data structure and discover insights

Steps:

1. Discover structure:

   • get_models → List available data models
   • get_explores → Find relevant explores
   • get_dimensions + get_measures → See available fields

2. Build initial query:

   • Start with key dimensions (grouping fields)
   • Add relevant measures (aggregations)
   • Use query tool to execute

3. Refine and filter:

   • Apply filters to narrow scope
   • Add sorts to prioritize insights
   • Adjust limit for appropriate data size

4. Analyze results:

   • Look for patterns, outliers, trends
   • Identify follow-up questions
   • Iterate with new queries

2. Creating Visualizations

Goal: Build charts and dashboards

Steps:

1. Start with a query (see workflow above)

2. Choose visualization type:

   • Line charts: Time series trends
   • Bar/Column: Comparisons across categories
   • Pie/Donut: Proportional breakdowns
   • Scatter: Correlations between metrics
   • Table: Detailed data review

3. Generate URL or create Look:

   • query_url tool: Get shareable Looker URL with vis_config
   • make_look tool: Save as permanent Look in personal folder

4. Add to dashboard:

   • Use add_dashboard_element to add tiles
   • Configure vis_config for visual styling
   • See reference/visualizations.md for vis_config details

3. Building Dashboards

Goal: Create comprehensive dashboard with multiple tiles

Steps:

1. Plan dashboard structure:

   • Identify key metrics and KPIs
   • Group related visualizations
   • Consider filter dependencies

2. Create dashboard tiles:

   • Build query for each tile
   • Choose appropriate visualization
   • Use add_dashboard_element repeatedly

3. Configure interactivity:

   • Set up dashboard filters
   • Enable cross-filtering if needed
   • Test drill-down paths

4. SQL Generation

Goal: See underlying SQL for debugging or learning

Steps:

1. Build query using normal parameters
2. Use query_sql tool instead of query
3. Review generated SQL
4. Optimize or troubleshoot as needed

5. Health Monitoring

Goal: Monitor Looker instance health and performance

Available checks:

• health_pulse with actions:

  • check_db_connections: Verify database connectivity
  • check_dashboard_performance: Find slow dashboards
  • check_dashboard_errors: Identify broken dashboards
  • check_explore_performance: Find slow explores
  • check_schedule_failures: Check scheduled delivery failures
  • check_legacy_features: Identify deprecated features

• health_analyze: Calculate usage metrics

  • Actions: projects, models, explores
  • Parameters: timeframe (days), min_queries

• health_vacuum: Suggest unused resources for cleanup

  • Actions: models, explores
  • Find candidates for deletion based on usage

---

Looker MCP Integration

Available MCP Tools

The Looker MCP server (looker-toolbox) provides these tools:

Discovery Tools

• get_models - List all LookML models
• get_explores - List explores in a model
• get_dimensions - List dimensions in an explore
• get_measures - List measures in an explore
• get_filters - List available filters
• get_parameters - List parameters

Query Tools

• query - Execute query and get JSON results
• query_sql - Generate SQL without executing
• query_url - Generate Looker URL with optional vis_config

Content Creation

• make_look - Create saved Look in personal folder
• add_dashboard_element - Add tile to dashboard
• run_look - Execute existing Look by ID
• get_looks - Search for Looks

Development Tools

• get_projects - List LookML projects
• dev_mode - Switch to/from development mode
• create_project_file - Create LookML file
• update_project_file - Modify LookML file
• delete_project_file - Remove LookML file

Health & Monitoring

• health_pulse - Run health checks
• health_analyze - Analyze usage patterns
• health_vacuum - Find unused resources

Tool Call Pattern

All MCP tools are called with this naming convention:

mcp__looker-toolbox__<tool_name>

Example:

mcp__looker-toolbox__query
mcp__looker-toolbox__get_models
mcp__looker-toolbox__make_look

Filter Expressions

Filters use special syntax - see reference/filter-expressions.md for complete syntax.

Quick examples:

{
  "filters": {
    "orders.status": "completed,pending",     // Multiple values (OR)
    "orders.created_date": "7 days",          // Relative date
    "users.age": ">=21 AND <=65",             // Number range
    "products.name": "%shirt%",               // Contains (wildcard)
    "orders.total": "NOT NULL"                // Not null
  }
}

Important: Use not null (not -NULL) for null checks.

---

When to Use This Skill

Invoke this skill when:

• Building Looker queries programmatically
• Creating visualizations or dashboards
• Analyzing data through Looker explores
• Generating SQL from Looker models
• Creating or modifying Looks
• Working with LookML development
• Monitoring Looker instance health
• Troubleshooting query performance
• Understanding Looker data structure
• Teaching users how to use Looker

---

Reference Materials

For detailed information, see these reference files (loaded on-demand):

• reference/explores-guide.md - Complete guide to using explores interactively
• reference/filtering.md - All filter types, operators, and expressions
• reference/expressions-and-calcs.md - Expressions, table calculations, and custom fields
• reference/functions-reference.md - Complete function library with examples
• reference/formatting-guide.md - Number and time formatting patterns
• reference/visualizations-guide.md - Chart types and vis_config options
• reference/merged-results-guide.md - Merging results from multiple explores

These files are loaded on-demand when you need specific details.

---

Key Principles

1. Start with discovery: Always use get_models, get_explores, and field discovery tools first
2. Build iteratively: Start simple, add complexity as needed
3. Validate fields: Use get_dimensions/get_measures to confirm field names
4. Check filter syntax: Reference filter expressions guide for correct syntax
5. Test queries: Use query tool before creating permanent Looks
6. Monitor performance: Use health tools to identify issues early
7. Document decisions: Explain query logic and visualization choices

---

Common Patterns

Pattern: Multi-Step Analysis

1. get_models → find relevant model
2. get_explores → find relevant explore
3. get_dimensions + get_measures → discover fields
4. query with simple fields → validate data
5. query with filters/sorts → refine results
6. query_url with vis_config → create visualization
7. make_look → save for reuse

Pattern: Dashboard Creation

1. Plan: Identify 4-6 key metrics
2. For each metric:
   - Build query
   - Choose visualization
   - call add_dashboard_element
3. Test interactivity
4. Share dashboard URL

Pattern: Performance Troubleshooting

1. health_pulse check_explore_performance → find slow explores
2. query_sql → examine generated SQL
3. Optimize: Reduce fields, add filters, check joins
4. query → validate improvement

---

Getting Help

For specific topics, tell me what you need:

• "How do I use explores?" → loads explores-guide.md
• "Show me filter syntax" → loads filtering.md
• "How do I create table calculations?" → loads expressions-and-calcs.md
• "What functions are available?" → loads functions-reference.md
• "How do I format numbers/dates?" → loads formatting-guide.md
• "What chart types can I use?" → loads visualizations-guide.md
• "How do I merge results?" → loads merged-results-guide.md

I'll load the relevant reference material and guide you through the task.

##Examples

See the examples/ directory for common patterns:

• Query building workflows
• Dashboard creation examples
• Complex filter patterns
• Advanced calculation techniques