Skip to main content
Model Context Protocol

Your catalog, in every AI client you already use.

MCP (Model Context Protocol) is how external AI clients — Claude, ChatGPT, Gemini, Cursor, and your IDE — call search_products, quick_publish, and 90+ other tools against your live tenant catalog, with your permissions.

In-platform specialist agents in the Agent Control Center use the same governed catalog, with persistent insights, an approval queue, and Catalog Quality monitoring. MCP and ACC complement each other — ACC for autonomous operations inside PIM; MCP for conversational and IDE-integrated workflows.

90+Structured catalog tools
OAuthHosted sign-in for Claude & ChatGPT
Same IAMActions run as you in your tenant
The Agentic PIM differentiator

Traditional PIMs export CSVs for chatbots. CataZenta exposes named, typed tools on the same Agentic PIM layer as the UI, syndication workers, and specialist agents — not a separate shadow catalog.


Three ways to use catalog AI

ModeWhereBest for
MCP (this page)Claude, ChatGPT, Cursor, VS CodeConversational ops, IDE automation, ad-hoc tool chains
Specialist agentsAgent Control CenterScheduled runs, missions, insights, approval queue
Zen AIChat inside PIMInline assistant on product pages — same MCP tools as external clients

All three operate on the same governed catalog. Writes from MCP are auditable API calls; in-platform agent writes default to the approval queue.


What you can do

Use caseExample promptsExample tools
Catalog Q&A“Show products in Footwear missing SEO titles”search_products, get_product_attributes
Enrichment“List SKUs below 70% completeness in Brand X”find_products_needing_enrichment, analyze_product_completeness
Approvals“What agent proposals are pending review?”Query via ACC in PIM; use list_tasks for workflow gates
Syndication“Publish these IDs to Shopify and report failures”list_connections, quick_publish, get_export_job_items
Governance“What workflow tasks are open for my team?”list_tasks, get_my_tasks, list_workflows
Analysis“Compare these two products side by side”compare_products, get_product_completeness
Media“List assets in the Spring campaign folder”list_dam_folders, list_assets
Bulk ops“Create an export job for all active SKUs in category …”search_products, create_export_job, start_export_job

Copy-paste prompts and multi-step patterns → Agent workflows · Automation examples


How it works

Your AI client discovers tools from the MCP endpoint, calls them with structured arguments, and receives JSON the model can parse. Every tool maps to the same REST APIs the PIM UI uses — scoped to your tenant and roles.

Sandbox first

Use a non-production tenant when testing write tools (set_product_attribute, quick_publish, create_product). Hosted and local connectors both respect tenant boundaries.


Connection options

OptionBest forSetup
Hosted MCP (recommended)Claude.ai, ChatGPTAdd URL + OAuth sign-in — no local install
Local MCP connectorClaude Desktop, Cursor, VS Code, WindsurfOptional package from CataZenta onboarding

MCP URL: https://api.catazenta.com/mcp

Credentials and connector packages are provided when your tenant is enabled for MCP (same API host as the REST API).

Hosted MCP

Connect in the AI app with the URL above. Sign in when prompted — same identity as the PIM.

Local connector

Run the CataZenta MCP connector on your machine for Desktop and IDE clients.

Also supported

Any MCP-compatible client that supports Streamable HTTP or stdio transport.


Tool coverage

90+ tools across twelve capability areas — all returning JSON for the model to parse. Parameter names and types are defined in the MCP tool schema exposed to your client.

Catalog core

Products, variants, families, attributes, categories.

  • search_products
  • set_product_attribute
  • list_categories

Quality & analysis

Completeness, enrichment gaps, comparisons.

  • analyze_product_completeness
  • find_products_needing_enrichment
  • compare_products

Syndication

Channels, export jobs, quick publish.

  • list_connections
  • quick_publish
  • create_export_job

Operations

DAM, jobs, workflows, tasks, dashboard.

  • list_assets
  • create_task
  • get_dashboard_stats

Full catalog with every tool name → MCP tools reference


Authentication

OptionHow it works
Hosted (Claude / ChatGPT)Add MCP URL → OAuth sign-in → Bearer token on each request — same account and tenant as PIM
Local connectorUsername/password or API token from onboarding — prefer a dedicated integration user with least privilege

Hosted clients discover OAuth settings from the MCP endpoint automatically. Tokens are scoped to your organization; the assistant only sees data your roles allow — identical to REST API authentication.

Security practices

  • Use HTTPS only (https://api.catazenta.com)
  • Rotate passwords and tokens on the same schedule as other integrations
  • Control which AI apps may use write-capable tools in your organization
  • Use workflow tasks and ACC approvals for regulated catalog changes

Details → MCP authentication · Security


Agent workflow patterns

Pre-built tool chains for reliable automations — not one-off chat messages.

PatternTool chainGuide
Catalog auditfind_products_needing_enrichmentanalyze_product_completenesscreate_taskAgent workflows
Publish botlist_connectionsquick_publishget_export_job_itemsPublish workflow
Content agentget_productset_product_attributeget_product_completenessAutomation examples
Human gates for production

Require PIM workflow sign-off before setting update_product_status to active in regulated catalogs. In-platform agent proposals use the ACC approval queue; MCP writes are immediate API calls — design guardrails accordingly.


MCP vs Agent Control Center

MCP clientsAgent Control Center
InteractionConversational / IDE tool callsAutonomous runs, schedules, missions
PersistenceSession-driven; you orchestrate chainsInsights, approval queue, Catalog Quality
WritesNamed tool calls, API-auditableProposals → approve → apply
Best forAd-hoc ops, developer automationContinuous catalog intelligence at scale

Both complement each other on the same Agentic PIM platform → Agent Control Center · AI agents (MCP clients)


Get started

  1. Confirm MCP is enabled for your tenant (contact CataZenta if needed).
  2. Follow the MCP quickstart — hosted URL or local connector.
  3. Pick your client guide (Claude, ChatGPT, Cursor, …).
  4. Test read-only: “Search for 5 products in my catalog.”
  5. Test analysis: “Find products below 70% completeness in category …”
  6. Wire write tools only after sandbox validation.
  7. Combine with ACC for persistent findings and governed agent writes.

MCP URL: https://api.catazenta.com/mcp