Skip to content

NinjaCat Integration Guide

NinjaCat serves two critical roles in Seer's AI agent infrastructure:

  1. Consumption Layer — Where practitioners run agents and interact with AI workflows
  2. Data Connector — Bridge to marketing data sources (Search Console, Google Ads, GA4, etc.)

This guide explains how NinjaCat fits into the overall architecture and how to get the most out of it.

Understanding the Build vs Consume Split

Seer's agent infrastructure intentionally separates building from consuming:

Role Tooling Purpose
Builders Claude, IDE, OpenCode Create and QA agents with full flexibility
Practitioners NinjaCat Run agents with client data connections

Why this split?

  • Builders need power tools: version control, debugging, modular files, rapid iteration
  • Practitioners need simplicity: one interface, automatic data connections, no setup required

Agents are built once (by builders) and consumed everywhere (by practitioners via NinjaCat).

NinjaCat as Consumption Layer

What You Can Do

When using agents through NinjaCat:

  • Run workflows — Execute pre-built agents like content audits, search landscapes, keyword analysis
  • Connect client data — Automatically pull from connected marketing accounts
  • Get structured outputs — Receive markdown files, outlines, and reports ready for refinement
  • Iterate with chat — Ask follow-up questions and refine outputs conversationally

Typical Workflow

1. Open NinjaCat
2. Select the agent (e.g., "Content Audit")
3. Choose client/project (data auto-connects)
4. Run the workflow
5. Review output in chat
6. Export as markdown → Google Doc
7. Refine and deliver

Tips for Best Results

Provide Context

The more specific your input, the better the output. Include:

  • Client name and project
  • Specific URLs or keyword sets
  • Business goals or constraints
  • Any prior analysis to reference

Use the Chat

Don't treat agents as one-shot tools. After the initial output:

  • Ask clarifying questions
  • Request alternative perspectives
  • Drill into specific findings
  • Ask for format changes

Markdown → Google Doc Workflow

For deliverable-ready outputs:

  1. Export agent output as .md file
  2. Upload to Google Drive
  3. Right-click → "Open with" → Google Docs
  4. Auto-formats with headings, tables, bullets
  5. Apply Seer brand template if needed

NinjaCat as Data Connector

Available Data Sources

NinjaCat connects to marketing platforms that would otherwise require separate authentication:

Platform Data Available Used By
Google Search Console Rankings, impressions, clicks, CTR SEO agents
Google Analytics 4 Sessions, conversions, user behavior Analytics agents
Google Ads Campaigns, spend, conversions, keywords PDM agents
Microsoft Ads Campaign performance, audience data PDM agents
Meta Ads Social campaign performance PDM agents
SeerSignals SERP snapshots, paid media daily data All divisions

How Data Connection Works

  1. Account is connected once in NinjaCat admin
  2. Agents reference data through natural language queries
  3. NinjaCat translates to appropriate API calls
  4. Results flow into agent context automatically

You don't need to: - Authenticate separately for each platform - Write SQL or API queries - Export/import CSV files - Manage data refresh schedules

Data Freshness

Source Typical Lag Notes
Google Search Console 2-3 days Google's standard delay
Google Analytics 4 Same day Real-time for most metrics
Google Ads Same day Conversion data may lag 1-2 days
SeerSignals Daily refresh SERP snapshots updated overnight

When to Use NinjaCat vs Other Tools

Use NinjaCat When:

  • ✅ Running established workflows with client data
  • ✅ You need automatic authentication to marketing platforms
  • ✅ Producing deliverables that need client-specific data
  • ✅ Working with non-technical team members
  • ✅ Standard workflows that don't need customization

Use Claude/OpenCode When:

  • ✅ Building or modifying agent definitions
  • ✅ Debugging agent behavior
  • ✅ Experimental or one-off analysis
  • ✅ Working with code, not just content
  • ✅ Need full control over prompts and context

Decision Framework

Need client marketing data?
  YES → NinjaCat (data connections built in)
  NO → Either works

Building/debugging an agent?
  YES → Claude/OpenCode (better dev tools)
  NO → NinjaCat (simpler interface)

Standard workflow for client deliverable?
  YES → NinjaCat (purpose-built)
  NO → Evaluate based on task

Common Questions

"Can I customize agents in NinjaCat?"

Limited customization is available through conversation. For structural changes to how an agent works, builders need to modify the agent definition in Claude/OpenCode, then deploy to NinjaCat.

"Why doesn't my agent have access to [data source]?"

Data sources must be: 1. Connected in NinjaCat admin (one-time setup) 2. Associated with the specific client/project 3. Referenced in the agent's design

If a data source isn't available, contact the Innovation team to check connection status.

"Can I use the same agent for different clients?"

Yes. Agents are client-agnostic. When you select a client/project in NinjaCat, the agent automatically pulls data for that context.

"What if NinjaCat is down?"

For critical work, you can: 1. Use Claude/OpenCode directly (no client data auto-connection) 2. Export data from platforms manually and provide as context 3. Focus on analysis that doesn't require real-time data

Best Practices

For Practitioners

  1. Start with the right agent — Use /utils:commands equivalent in NinjaCat to see available workflows
  2. Provide complete context — Client, project, date range, specific focus areas
  3. Iterate in chat — Refine outputs before exporting
  4. Use markdown export — Cleanest path to deliverable-ready documents
  5. Report issues — If an agent behaves unexpectedly, flag it for builders

For Builders

  1. Test in NinjaCat — Before releasing, verify agent works with real client data
  2. Document data requirements — Clear about what platforms/data the agent needs
  3. Handle missing data gracefully — Agents should work even if some data sources unavailable
  4. Consider the practitioner — They shouldn't need to understand implementation details

Last updated: January 2026