Skip to main content

SAP Integrates Claude into Joule: Model Context Protocol Opens the Way for Enterprise AI Agents

Ilustrační obrázek
SAP, in collaboration with Anthropic, announced the integration of the Claude model into its AI platform Joule. A key role is played by the Model Context Protocol (MCP) — an open standard that allows AI agents to communicate across enterprise systems without the need for custom integration for each individual tool. For the hundreds of thousands of companies using SAP, this means a fundamental shift: Claude-powered agents will be able to perform specific tasks in finance, HR, procurement, and supply chains — within existing workflows, not in place of them.

Listen to this article:

What's Happening: Claude Enters the SAP World

SAP and Anthropic announced a strategic integration on May 20, 2026 — the Claude model becomes part of the SAP Business AI Platform and the Joule assistant. According to SAP CTO Philipp Herzig, Claude will function both as a foundational model within the platform and as an extension of Joule's orchestration for agentic tasks.

This is not just another generic "AI in ERP." The integration uses the Model Context Protocol (MCP) — an open standard that Anthropic introduced at the end of 2024 and which has quickly become an industry norm for communication between AI agents and external systems. SAP has been heavily supporting MCP since its TechEd 2025 conference, and the SAP HANA Cloud database platform has had full MCP support since early 2026.

Model Context Protocol: Esperanto for AI Agents

MCP works as a universal translator between AI models and enterprise systems. Instead of developers writing separate integrations between Claude and every SAP application — S/4HANA, SuccessFactors, Ariba — they use a single standardized protocol understood by all MCP-compatible systems.

In practice, this means a Claude agent can:

  • Retrieve customer data from SAP S/4HANA
  • Pull employee records from SAP SuccessFactors
  • Check inventory status in the supply chain system
  • Trigger approval workflows in SAP Ariba

All using the same protocol, without custom code for each integration. And importantly — MCP is not a proprietary SAP or Anthropic technology. It's an open standard already supported by other platforms, including SAP Integration Suite, meaning agents can also communicate with external third-party systems.

How Claude and Joule Collaborate in Practice

Joule, SAP's assistant launched in 2023, has so far primarily functioned as an interface for queries and recommendations. Now, thanks to Claude, it gains agentic capabilities — the ability to independently perform specific actions.

According to the SAP Architecture Center documentation, when processing a request, Joule analyzes three categories of information: the Scenario Catalog (metadata of all available scenarios and functions across SAP cloud applications), the Knowledge Catalog (corporate and customer knowledge base), and user context including roles, permissions, and conversation history.

SAP provides a concrete example: a treasury manager can ask Joule to prepare a presentation for the CFO ahead of a meeting with the bank. Within minutes, Joule delivers a finished presentation with current data, analysis, and notes on financial risks — a task that previously required hours of manual work.

What It Means for Companies — and What to Watch Out For

The integration of Claude into SAP promises a dramatic acceleration of routine business processes across finance, HR, procurement, and logistics. For managers, however, it also raises new questions:

Who Manages MCP Integrations?

When a procurement agent depends on live MCP connections between S/4HANA, Ariba, SuccessFactors, and external supplier portals, a failure of a single integration doesn't just stop a workflow — it can block the entire purchasing decision-making process. Before deploying agentic automation, it's essential to clarify with system integrators who bears responsibility for testing, monitoring, and supporting the integrations.

Single-Vendor Dependency

SAP presents the partnership as an open ecosystem that gives customers more choice. However, deeper integration with Claude also means deeper dependency on Anthropic's product roadmap, pricing, and availability. Tighter model coupling usually means tighter vendor lock-in — and that has implications for costs, risks, and negotiating position.

AI That Recommends vs. AI That Acts

The difference between a system that makes recommendations and a system that independently performs actions is enormous. Especially in regulated industries where audit trails and human accountability are non-negotiable. SAP emphasizes that role-based permissions determine what agents can access — but the question remains at exactly what point in the process human oversight occurs.

SAP, Anthropic, and the Competitive Landscape

This integration is not an isolated event. It fits into a broader trend of tech giants racing to dominate agentic AI in enterprises:

  • Microsoft is betting on Copilot across Dynamics 365 and Microsoft 365 with OpenAI models
  • Google offers Gemini within Google Cloud and Workspace
  • Oracle is building its own AI agents for Fusion Cloud
  • Salesforce is expanding Einstein AI with agentic capabilities

SAP and Anthropic are taking a different path — instead of developing their own proprietary model, they are opening the platform to an external model and handling connectivity through the open MCP standard. This is a sensible strategy: SAP doesn't need to compete in the race for the most powerful language model, while at the same time providing customers with cutting-edge AI from Anthropic, which over the past year has introduced the Claude Opus 4.7 and Claude Sonnet 4 models with excellent results in enterprise deployment.

Anthropic and SAP are also planning to jointly develop custom agents for key industries: public sector, healthcare, education, pharmaceuticals, and energy. This suggests it's not a one-off integration but a long-term strategic collaboration.

The Czech and European Dimension

This news has a direct impact on Czech companies. SAP is the backbone information system in the Czech Republic — used by most large enterprises, from Škoda Auto to ČEZ to banks and insurance companies. Integrating Claude into Joule opens the door to agentic AI without the need to leave the existing SAP environment.

From a European regulatory perspective, it's notable that both SAP (a German company) and Anthropic (a U.S. company, but with an emphasis on safety and transparency) declare compliance with the requirements of the EU AI Act. For European companies that must consider compliance, the combination of a proven ERP system with a model known for its safety focus (Anthropic's Constitutional AI) is a relevant choice.

The question of Czech localization remains — Joule currently works primarily in English and German. Claude handles Czech reasonably well, but full-scale deployment in Czech companies will require official language support. SAP is nevertheless gradually expanding Joule's language variants, and with growing emphasis on the Central European market, Czech can be expected as well.

When will the Claude integration into SAP Joule be available to customers?

SAP has not yet announced a specific general availability date. The integration was announced as a strategic partnership with phased rollout. The first deployments are expected in pilot mode during 2026 for selected customers and industries.

Will the Claude integration into SAP Joule be charged separately?

SAP and Anthropic have not yet disclosed a pricing model. It can be expected that Claude-powered agentic features will be part of higher-tier SAP Business AI Platform plans or a separate subscription. Companies should anticipate additional costs beyond standard SAP licenses — similar to Microsoft Copilot for Dynamics 365.

Can I use Claude agents in SAP Joule alongside other AI models?

Yes, SAP emphasizes the platform's openness. Joule orchestrates multiple models available through SAP AI Foundation and the Generative AI Hub. Claude is a new, significant choice — not the only option. Companies can thus combine different models based on specific needs, budget, and regulatory requirements.

X

Don't miss out!

Subscribe for the latest news and updates.