OpenDXP MCP Bundle

Give your AI assistant safe, governed access to your OpenDXP content.

A bundle by PRinguin for OpenDXP. The OpenDXP MCP Bundle connects OpenDXP to the Model Context Protocol (MCP) — the open standard that AI assistants such as Claude, GitHub Copilot, and Cursor use to talk to external systems. Once installed, your team can create, search, update, and publish Data Objects, Assets, Documents, and Shared Translations using plain-language prompts, while you stay in full control of exactly what the AI is allowed to see and do.

ℹ️ This is a public overview repository. The source code is distributed privately to our customers. If the bundle looks like a fit, see Getting access and get in touch — we'll get you set up.

Why it matters

Content teams spend hours on repetitive PIM/DXP work: creating product records, filling in localized fields, maintaining translation catalogs, building landing pages block by block, exporting data for reports. The OpenDXP MCP Bundle lets an AI assistant do that work inside your OpenDXP instance — driven by natural language, executed through your existing permission model.

• Talk to your content. "Find every product whose name contains 'merino' and list them with SKU and price." "Create an imprint page under /de with these two sections and the company logo."
• Automate the tedious parts. Bulk metadata edits, CSV exports, translation upkeep, page scaffolding — described once, executed in seconds.
• Stay in control. Nothing is exposed unless you explicitly whitelist it. Every action runs under a real, permission-bound OpenDXP user.

What it does

The bundle turns your OpenDXP instance into an MCP server. AI clients connect to it and automatically discover the tools and data you have chosen to expose.

See FEATURES.md for the full capability breakdown, and USE_CASES.md for concrete example prompts.

How it works

┌─────────────────┐        MCP         ┌────────────────────────┐        ┌──────────────┐
│  AI assistant   │  ───────────────▶  │   OpenDXP MCP Bundle    │  ───▶  │   OpenDXP    │
│ Claude / Copilot│  ◀───────────────  │  (MCP server endpoint)  │  ◀───  │  (your data) │
│    / Cursor     │   tools & data     │  whitelist + permissions│        │              │
└─────────────────┘                    └────────────────────────┘        └──────────────┘

1. You configure which classes, fields, and actions are exposed — a strict whitelist.
2. The AI client connects over STDIO (local) or authenticated HTTPS (remote).
3. Every request is checked against OpenDXP's own permission system before anything happens.
4. Every call is audit-logged with tool, object, user, timestamp, and result.

Governance & security first

This bundle is built for production OpenDXP environments, where "let the AI touch our content" only works if it is genuinely safe.

• Whitelist principle. Only classes you explicitly list are visible. Everything else is invisible to the AI.
• Field-level control. Readable, writable, and hidden fields are configured per class. A global blacklist guarantees passwords, tokens, and secrets are never returned — even if asked for directly.
• Permission-bound. Every operation runs as an authenticated OpenDXP user and respects element-, action-, and field-level permissions. If a user can't do it in the admin UI, they can't do it via MCP.
• Authenticated transport. The remote HTTP transport uses Bearer authentication against OpenDXP admin credentials and is designed to run over HTTPS only.
• Full audit trail. Every tool call is logged for review.

More detail in SECURITY.md.

System requirements

Full details, including a minimal configuration example, are in REQUIREMENTS.md.

Configuration in 30 seconds

Exposing a class is a few lines of YAML — this is all it takes to make a Product class searchable and readable by an AI assistant:

# config/packages/opendxp_mcp.yaml
opendxp_mcp:
  classes:
    Product:
      actions: [search, get, list]
      searchable: [name, sku]

Add create, update, csv_export, more fields, and locales as you grow comfortable. Nothing you don't configure is ever exposed.

Getting access

We love open source and we try to contribute to OpenDXP as much as possible. And we do offer free open source bundles for OpenDXP and Pimcore. However maintaining a bundle like this is expensive and therefor we need to compensate our efforts.

The bundle is distributed to licensed customers via a private GitHub repository. To evaluate or purchase it:

1. Get in touch — email [email protected] or use the contact form at www.pringuin.com.
2. Tell us a little about your setup: OpenDXP version, languages, expected volume, and which provider(s) you intend to use.
3. We'll grant your GitHub account access to the private package repository and walk you through installation.

Pricing is on request — reach out and we'll put together an offer that fits your project.

Frequently asked questions

Short answers to the most common questions — see FAQ.md for the full list.

• Is my data sent to the AI provider? Only the specific content the assistant requests for a task, and only from what you whitelisted. You control the exposure surface.
• Can it accidentally delete or overwrite content? Only if you enable those actions for that class, and only if the connected user has permission. Destructive actions can be left out entirely.
• Do we need to build integrations for each AI tool? No. MCP is a standard — any compliant client connects to the same server.

About PRinguin

PRinguin GbR builds and maintains extensions and integrations for and websites with OpenDXP and Pimcore. Learn more at www.pringuin.de.

Independent community bundle for OpenDXP. Not affiliated with OpenDXP, Pimcore GmbH nor with Claude, GitHub Copilot or Cursor.

License

The bundle is licensed under GPL-3.0-or-later. See LICENSE.md.