MadAI Support Ticket Investigator

Investigate MadKudu support tickets systematically using available MCP tools.

Investigation Workflow

Follow these steps in order for thorough investigation:

Step 1: Fetch Ticket Details

Extract ticket information based on source:

Zendesk tickets:

mcp__MadAI_Prod__get_zendesk_ticket(ticket_url)

Jira tickets:

mcp__MadAI_Prod__get_jira_ticket(issue_key)  # e.g., "SUPPORT-123"

Identify from ticket:

Tenant ID (look for tenant mentions, domain, or company name)
Issue description and symptoms
Timeline (when did the issue start?)
Any error messages or screenshots

Step 2: Identify Tenant

If tenant ID not explicit in ticket, search by domain:

mcp__MadAI_Prod__get_argo_data(endpoint="tenant", params={"domain": "example.com"})

Or search by user email:

mcp__MadAI_Prod__get_argo_data(endpoint="users/search", params={"email": "user@example.com"})

Get full tenant details:

mcp__MadAI_Prod__get_argo_data(endpoint="tenant/{tenant_id}")

Step 3: Health Checks

Run health checks in parallel when possible:

Tenant active status:

mcp__MadAI_Prod__check_tenant_active(tenant_id)

Connector connectivity:

mcp__MadAI_Prod__check_connector_connectivity(tenant_id)

Fill rate (scoring health):

mcp__MadAI_Prod__get_fill_rate_data(tenant_id, look_back_hours=24)

Process status:

mcp__MadAI_Prod__get_pull_processes_status(tenant_id, look_back_hours=24)
mcp__MadAI_Prod__get_push_processes_status(tenant_id, look_back_hours=24)
mcp__MadAI_Prod__get_data_workers_status(tenant_id, look_back_hours=24)

Step 4: Dig Deeper (as needed)

Check connector details:

mcp__MadAI_Prod__get_argo_data(endpoint="tenant/{tenant_id}/connectors/{connector_name}")
# connector_name: salesforce, hubspot, marketo, etc.

Query Datadog logs for errors:

mcp__MadAI_Prod__query_datadog_logs(query="@tenant:{tenant_id} status:error", look_back_hours=72)

Query tenant databases:

# Redshift (analytics data)
mcp__MadAI_Prod__execute_redshift_query(tenant_id, query="SELECT ...")

# Postgres - emback (main app data)
mcp__MadAI_Prod__execute_postgres_query(tenant_id, query="SELECT ...", database="emback")

# Postgres - honk (playbooks data)
mcp__MadAI_Prod__execute_postgres_query(tenant_id, query="SELECT ...", database="honk")

Check tenant config in GitHub:

mcp__MadAI_Prod__get_github_tenant_config_file_content(file_path="path/to/config")

Step 5: Query Knowledge Bases

Search for similar issues or solutions:

Product questions:

mcp__MadAI_Prod__answer_product_question(question="How does X feature work?")

Engineering how-to:

mcp__MadAI_Prod__answer_engineering_question(question="How to debug Y issue?")

Step 6: Produce Investigation Report

Structure findings as:

## Investigation Summary

**Ticket:** [ticket ID/URL]
**Tenant:** [tenant_id] - [company name]
**Issue:** [brief description]

## Findings

### Health Status
- Tenant Active: ✅/❌
- Connectors: [status]
- Fill Rate: [percentage]
- Processes: [status]

### Root Cause Analysis
[What's causing the issue based on evidence]

### Evidence
- [Log entries, error messages, data points]

## Recommendations

1. [Action item]
2. [Action item]

## Next Steps
[What should happen next - escalate, fix, follow up]

Common Investigation Patterns

Scoring Issues

Check fill rate data
Verify connector connectivity
Query pull/push process status
Check Datadog for scoring errors

Connector Problems

Check connector connectivity
Get specific connector details from Argo
Query Datadog for connector-specific errors
Verify credentials/config in tenant settings

Data Sync Issues

Check data workers status
Query pull processes
Check Redshift for data freshness
Look for sync errors in Datadog

Missing Scores/Signals

Verify tenant is active
Check if model is deployed (tenant config)
Query Postgres for signal configuration
Check fill rate trends over time