Concept card

MCP — Model Context Protocol in SAP

As of 2026-07-24T14:00:00Z

Edited by Cédric-Olivier Mâry, sole editor and publication director — 27 years in the SAP ecosystem. How these pages are produced, and where automation is used.

What is MCP — Model Context Protocol in SAP?

How the Model Context Protocol concretely shows up in SAP’s Joule stack: a single versioned JSON-RPC protocol lets one Joule session query an S/4HANA MCP server and a Datasphere MCP server simultaneously, with neither server aware of the other. For the adopt-or-not decision, see C105.

The Model Context Protocol (MCP) is an open protocol, built on JSON-RPC 2.0, that standardises how a large language model or AI agent connects to external tools, data, and workflows. Anthropic published MCP openly in November 2024, and the specification was adopted quickly across the industry because it solves a problem every enterprise AI programme eventually hits: without a shared protocol, each new tool integration is a bespoke connector, and a multi-tool agent becomes a tangle of one-off adapters that is expensive to maintain, hard to audit, and brittle whenever an underlying API changes. MCP replaces that web of point-to-point wiring with one interface that both the model-hosting client and the tool-exposing server implement once.

In SAP terms, MCP is the connective tissue between Joule — SAP's AI copilot — and the wider landscape of SAP services on Business Technology Platform: S/4HANA, Datasphere, SAP Analytics Cloud, SuccessFactors, and third-party systems a customer has wired in. Rather than a developer hard-coding calls to individual OData services or BAPIs for every scenario, an MCP-compliant agent discovers what a server offers at runtime and calls it through a standard envelope, so the model reasons over structured tool descriptions instead of undocumented API quirks.

How it is put together

Why it matters

Published by Anthropic in late 2024 and adopted industry-wide — the standardisation layer that makes multi-tool Joule agents maintainable and auditable instead of fragile.
The clear client-server split (server owns and exposes capabilities, client discovers and calls on the model's behalf) clarifies who builds what in an SAP integration project.
Three distinct capability primitives give a structured vocabulary for scoping an MCP server rather than treating it as one generic connector type.

Key points

MCP — Model Context Protocol in SAP is mastered only when it changes a named buyer decision.
Start with the semantic contract and control model before demonstrating the tool.
Use current SAP, analyst, study, KG, and news signals as evidence, not decoration.
Separate verified facts from directional trends and modeled assumptions.
Define owner, metric, threshold, support path, and rollback before scaling.
For AI use cases, measure reliability, cost, latency, safety, and human validation.
Leave a reusable operating asset: memo, checklist, control table, and exception log.
A premium answer is short, trade-off explicit, and defensible in a steering committee.

Terms used on this page

Decision owner: The accountable person who accepts the trade-off and funds the next action.
Semantic contract: The shared definition of business terms, metrics, entities, and access rules used by tools and teams.
Control plane: The layer that applies policy, access, lineage, monitoring, and escalation across the operating model.
Evidence grade: A label that separates verified fact, directional signal, modeled assumption, and field observation.
Adoption metric: The measurable behavior proving that the concept changed actual work after go-live.
Agent reliability: The consistency, cost, safety, and policy compliance of an agent across repeated runs.
Decision owner: The accountable person who accepts the trade-off and funds the next action.
Semantic contract: The shared definition of business terms, metrics, entities, and access rules used by tools and teams.

Sources

Full card available to members. What the full card adds: the full decision framework · the SAP vs Snowflake / Databricks / Fabric comparison · the common pitfalls and their fix · the cheat sheet · the architecture schemas · the code blocks · the facts worth quoting.

Open in the app →