AI Agents Overview - AnomalyArmor

Three ways to access AnomalyArmor AI agents: Web Agent, MCP Server, and Python SDK, all sharing the same 40+ tools

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)

Go to any asset’s Intelligence tab
Click Ask Agent
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"
    }
  }
}

Skills (Any Agent)

Install the 14 /armor:* skills into Claude Code, Cursor, Windsurf, Codex, OpenCode, or any of 40+ supported agents using the skills.sh CLI:

npx skills add anomalyarmor/agents

Source: github.com/anomalyarmor/agents

Example: Morning Health Check

Web Agent
MCP (Claude Code)

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.

$ 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
MCP (Claude Code)

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.

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
MCP (Claude Code)

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?

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.

Common Questions

Do I need Claude to use AnomalyArmor’s AI agents?

No. The Web Agent runs in your browser with no install, and the MCP server works with any MCP-compatible client, including Cursor, Windsurf, Codex, OpenCode, and Claude Code. The /armor:* skills install into 40+ supported agents via skills.sh.

What’s the difference between the Web Agent, MCP server, and Python SDK?

All three call the same underlying tools. The Web Agent is built into the Intelligence page with zero setup. The MCP server lets your IDE agent manage observability without switching apps. The Python SDK is for scripts, notebooks, and CI/CD automation.

How do I install the /armor skills in my IDE?

Run npx skills add anomalyarmor/agents and the 14 skills install into any supported agent (Claude Code, Cursor, Windsurf, Codex, OpenCode, and more). Source is at github.com/anomalyarmor/agents.

Can the agent make changes to my monitoring setup, or is it read-only?

The agent can create freshness schedules, metrics, validity rules, tags, and alert rules. For any action that modifies data, it shows the plan and asks for confirmation before executing. Tool calls are always visible so you can verify what ran.

Does the agent have access to my actual data rows?

No. The agent queries AnomalyArmor’s metadata (schemas, freshness timestamps, monitoring results, lineage), not your row-level data. See Query Gateway for how AnomalyArmor reads customer warehouses without exfiltrating rows.

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

​Why AI Agents?

​Available Skills

​Core Skills

​Extended Skills

​Quick Start

​Web Agent (No Setup)

​MCP Server (IDE)

​Skills (Any Agent)

​Example: Morning Health Check

​Example: Investigating an Alert

​Example: Setting Up Monitoring

​Common Questions

​Do I need Claude to use AnomalyArmor’s AI agents?

​What’s the difference between the Web Agent, MCP server, and Python SDK?

​How do I install the /armor skills in my IDE?

​Can the agent make changes to my monitoring setup, or is it read-only?

​Does the agent have access to my actual data rows?

​Next Steps