AI Agents Overview

Manage data observability through natural language, in the browser or your IDE

AnomalyArmor gives you three ways to interact with your data platform using natural language:

Access Point	Where	Best For
Web Agent	Built into the Intelligence page	Diagnosing issues while looking at your data. No setup required.
MCP Server	Claude Code, Cursor, any MCP client	Managing observability from your IDE without switching apps.
Python SDK	Scripts, notebooks, CI/CD	Automating checks, building workflows, programmatic access.

All three share the same underlying tools and API. The difference is where you interact.

Why AI Agents?

Instead of clicking through dashboards, just ask:

"What tables are unhealthy right now?"
"Add freshness monitoring to all gold tables"
"What changed upstream that could cause NULLs in silver_orders?"

Your AI assistant queries APIs, chains tools together, and interprets results for you.

Available Skills

Core Skills

Skill	Description	Example
/armor:status	Health summary	”Is my data healthy?”
/armor:alerts	Query and manage alerts	”What alerts fired yesterday?”
/armor:ask	Natural language Q&A	”What tables contain PII?”
/armor:analyze	Trigger AI analysis	”Analyze the finance schema”
/armor:connect	Connect a data source	”Connect my Snowflake warehouse”
/armor:monitor	Set up monitoring	”Monitor freshness for orders hourly”

Extended Skills

Skill	Description	Example
/armor:quality	Data quality checks	”Add null check for email column”
/armor:tags	Data classification	”Tag this table as PII”
/armor:investigate	Root cause analysis	”Why is this table stale?”
/armor:lineage	Data dependencies	”What depends on this table?”
/armor:profile	Table statistics	”Profile the orders table”
/armor:coverage	Monitoring coverage	”What tables have no alerts?”

Quick Start

Web Agent (No Setup)

1. Go to any asset’s Intelligence tab
2. Click Ask Agent
3. Start chatting

MCP Server (IDE)

# Claude Code (one command)
claude mcp add anomalyarmor --transport http https://mcp.anomalyarmor.ai/mcp

# Cursor (add to mcp.json)
{
  "mcpServers": {
    "anomalyarmor": {
      "url": "https://mcp.anomalyarmor.ai/mcp"
    }
  }
}

Sign in through your browser when prompted. See MCP Server setup for details.

Example: Morning Health Check

Web Agent

You: What should I prioritize this morning?

Agent: Here's your priority list:

       CRITICAL:
       - gold.fact_orders is 6 hours stale (SLA: 2h)
         Upstream staging.orders_raw is also stale.

       HIGH:
       - Schema drift: discount_pct column removed from
         gold.fact_orders on Tuesday
       - Null rate on gold.dim_customer.email spiked to 12.4%

       MEDIUM:
       - 2 validity rules failing on gold.dim_customer

       All other monitored tables are healthy.

MCP (Claude Code)

$ claude

You: /armor:status

Claude: Let me check your data health...

[Tool: health_summary]

Overall: NEEDS ATTENTION

3 issues found:
- 1 critical freshness violation (gold.fact_orders, 6h overdue)
- 1 schema drift alert (column removed from gold.fact_orders)
- 1 data quality anomaly (null rate spike on dim_customer.email)

45 other assets are healthy. Want me to investigate any of these?

Example: Investigating an Alert

Web Agent

You: I got a freshness alert on the orders table. What's happening?

Agent: gold.fact_orders is 6 hours overdue (SLA: 2 hours).

       I checked the upstream lineage:

         raw.events (FRESH)
           > staging.orders_raw (STALE, 8h overdue)
             > gold.fact_orders (STALE, 6h overdue)

       The issue originated in staging.orders_raw. 7 downstream
       tables are also affected:
       - gold.fact_payments
       - gold.fact_refunds
       - analytics.order_summary
       - reporting.daily_revenue
       ... and 3 more

       Recommended: Check the ETL pipeline that populates
       staging.orders_raw.

MCP (Claude Code)

You: The orders table is stale. Why?

Claude: Let me investigate...

[Tool: check_freshness > gold.fact_orders: STALE, 6h overdue]
[Tool: get_lineage > upstream: staging.orders_raw]
[Tool: check_freshness > staging.orders_raw: STALE, 8h overdue]

The orders table is stale because its upstream dependency
staging.orders_raw stopped updating 8 hours ago. The issue
started upstream.

7 downstream tables are affected. Want me to list them?

Example: Setting Up Monitoring

Web Agent

You: Set up hourly monitoring for all gold fact tables

Agent: Found 4 fact tables in the gold schema. Created freshness
       schedules:

       - gold.fact_orders: hourly, 2h SLA
       - gold.fact_payments: hourly, 2h SLA
       - gold.fact_refunds: hourly, 2h SLA
       - gold.fact_subscriptions: hourly, 2h SLA

       Want me to also add null_percentage metrics on key columns?

MCP (Claude Code)

You: Monitor all gold fact tables for freshness, hourly

Claude: I'll set those up now.

[Tool: search_tables > 4 fact tables in gold schema]
[Tool: create_freshness_schedule > gold.fact_orders: hourly, 2h SLA]
[Tool: create_freshness_schedule > gold.fact_payments: hourly, 2h SLA]
[Tool: create_freshness_schedule > gold.fact_refunds: hourly, 2h SLA]
[Tool: create_freshness_schedule > gold.fact_subscriptions: hourly, 2h SLA]

Done. 4 freshness schedules created. You'll get alerts if any
table doesn't update within 2 hours.

Next Steps

Web Agent

Chat with the agent in your browser

MCP Server

Connect your IDE

All Skills

Browse available skills

Python SDK

Programmatic access