Germany is writing MCP, A2A, and AG-UI into its federal IT architecture. The Deutschland-Stack 2.0, released by the Federal Ministry for Digital and State Modernization (BMDS) after a second public consultation closing February 15, 2026, mandates these open protocols as the interoperability standard for AI agents across all levels of German government. Standards must be finalized by March 31, 2026. Concrete deployments for federal, state, and municipal governments are targeted for 2028. And an agentic AI platform built on these principles already won the “Best Use of AI in Government Services” prize at the World Government Summit in Dubai on February 5, 2026.
This is not another aspirational whitepaper. The first version of the Deutschland-Stack, published in October 2025, was criticized as vague by the Open Source Business Alliance and industry groups. Version 2.0 responds with specific protocol names, architecture principles, and a deployment timeline that ties to a legal mandate: by 2029, German citizens gain a legal right to digital administrative services.
The Four Protocols Germany Chose
The Deutschland-Stack 2.0 specifies four protocols for its agentic AI layer. If you have been following the AI agent protocol wars, these names will be familiar.
Model Context Protocol (MCP)
MCP, originally created by Anthropic and now governed by the Agentic AI Foundation under the Linux Foundation, handles how AI agents access tools and data sources. The Deutschland-Stack describes it as the “USB-C for AI applications.” MCP uses JSON-RPC 2.0 to standardize how an agent reads a database, queries an API, or triggers an action in a government IT system.
For the German government, this means a single integration standard instead of hundreds of bespoke connectors between AI systems and the patchwork of federal, state, and municipal IT backends. A case worker’s AI assistant in Bavaria and one in Schleswig-Holstein would use the same protocol to query their respective permit databases.
Agent-to-Agent (A2A) and Agent Network Protocol (ANP)
A2A, launched by Google, and ANP handle communication between autonomous agents. In the Deutschland-Stack vision, government AI agents do not just respond to queries. They coordinate. An agent processing a building permit in one municipality could automatically request environmental impact data from another agency’s agent, then pass the compiled assessment to a third agent handling the final approval.
This agent-to-agent coordination is what the BMDS calls a “paradigm shift” from digital forms to proactive, networked government systems. The administration would not just digitize existing paper processes; it would restructure them around agent workflows.
AG-UI: Agent-User Interaction Protocol
AG-UI governs how citizens and civil servants interact with AI agents. This is the front-end protocol, defining how an agent presents options, requests confirmation, and hands off to a human when it hits the limits of its authority. The Deutschland-Stack explicitly requires that “legal responsibility and the final decision clearly remain with human case workers,” making AG-UI the enforcement layer for human oversight.
What Is Already Running: AI Agents in German Government
This is not all future tense. Several German agencies already deploy agentic AI systems that foreshadow the Deutschland-Stack’s broader vision.
Hamburg: Hydrogen Infrastructure Permitting
The Free Hanseatic City of Hamburg uses an agentic AI platform for approving hydrogen core network lines. The system analyzes extensive application documents in hours instead of the weeks or months required for manual review. It structures content automatically, checks for completeness, identifies gaps in evidence and expert opinions, and generates decision proposals based on legal analysis.
This is the platform that won the Dubai prize. Federal Digital Minister Karsten Wildberger stated that the technology “reduces the duration of processes that currently take months to just a few days.” The ministry plans to release the technology as open-source code for reuse across the country.
North Rhine-Westphalia: BImSchG Emission Permits
NRW is integrating AI agent modules for permit approval processes under the Federal Emission Control Act (BImSchG). These processes involve hundreds of pages of technical documentation, environmental assessments, and legal requirements that previously required months of expert review. The AI agents pre-process and validate documentation before human reviewers make the final call.
Federal Road Authority: Infrastructure Planning
A third implementation runs with the Federal Road Authority (Autobahn GmbH), applying similar agentic AI approaches to highway and infrastructure planning approvals. The common thread: complex, document-heavy approval processes where AI agents handle analysis and preparation while humans retain decision authority.
Architecture Principles: API-First, Open Source, Made in EU
The Deutschland-Stack 2.0 does not just pick protocols. It sets architecture rules that shape how these protocols get implemented.
API-First: Every service in the stack must expose a standardized API. No more screen-scraping legacy systems or building point-to-point integrations. This is the foundation that makes MCP integration possible at scale.
Open Source as Default: After criticism that version 1.0 could enable “sovereignty washing” (slapping a European label on proprietary solutions), version 2.0 mandates open source as the primary approach for in-house development. The Open Source Business Alliance (OSBA) and the Free Software Foundation Europe (FSFE) both contributed to the consultation.
DevSecOps by Default: Security and compliance are baked into the development pipeline, not bolted on afterward. Given the EU AI Act requirements that Germany is implementing through the KI-MIG, this is not optional.
Made in EU: The procurement framework prioritizes European providers meeting sovereignty criteria. The Deutschland-Stack explicitly positions itself as an alternative to dependence on US hyperscaler cloud platforms, though it stops short of excluding them entirely.
These principles address a real problem. Germany’s public IT is famously fragmented. Sixteen states, thousands of municipalities, and dozens of federal agencies each run their own systems. The Marketplace Germany Digital and the German Administrative Cloud portal serve as the delivery vehicles for the standardized stack.
The Timeline: Standards in Weeks, Deployment by 2028
The Deutschland-Stack 2.0 sets hard deadlines, which is unusual for German government IT projects.
| Milestone | Date |
|---|---|
| Second consultation round closes | February 15, 2026 |
| Standards and governance framework finalized | March 31, 2026 |
| BMDS presents updated overall picture | Autumn 2026 |
| Concrete offerings for all government levels | 2028 |
| Legal right to digital administrative services | 2029 |
The March 31 deadline is particularly significant. That is when the protocol standards, governance structures, and certification requirements must be locked in. Everything that follows builds on that foundation. State Secretary Markus Richter called the consultation process “an important step toward interoperable administration.”
The 2029 legal right to digital services creates the pressure valve. When citizens can legally demand digital interactions with their government, every agency needs compatible infrastructure. The Deutschland-Stack is meant to be that infrastructure.
What This Means for Companies Building AI Agents
If you sell AI tools or build agent systems, the Deutschland-Stack creates both an opportunity and a compliance surface.
The opportunity: Germany’s public sector is one of the largest IT markets in Europe. A government-mandated stack built on open protocols means that any company whose products support MCP and A2A can potentially serve the German public sector, without being locked out by proprietary standards. The open-source mandate lowers the barrier further.
The compliance surface: If your AI agents interact with German government systems, they will need to speak MCP and A2A. Your AG-UI implementation must preserve human oversight. Your infrastructure must meet the “Made in EU” sovereignty criteria, or you need a European partner who does. And your systems must comply with the KI-MIG, Germany’s implementation of the EU AI Act, which assigns the Bundesnetzagentur as the central AI regulator with fines up to EUR 35 million or 7% of global turnover.
Companies already building on MCP and A2A are in a strong position. The protocols are the same ones the private sector already uses. The Deutschland-Stack is not inventing new standards; it is adopting the ones that won the protocol standardization race and mandating them for government use.
Frequently Asked Questions
What is the Deutschland-Stack 2.0?
The Deutschland-Stack 2.0 is Germany’s federal technology platform for digitalizing all levels of public administration. Updated in February 2026, it specifies MCP, A2A, ANP, and AG-UI as mandatory protocols for AI agent interoperability across federal, state, and municipal government systems, with standards due by March 31, 2026, and full deployment targeted for 2028.
Which AI agent protocols does the Deutschland-Stack use?
The Deutschland-Stack 2.0 adopts four protocols: Model Context Protocol (MCP) for connecting AI agents to tools and data, Agent-to-Agent Protocol (A2A) and Agent Network Protocol (ANP) for autonomous inter-agent communication, and AG-UI (Agent-User Interaction Protocol) for governing how citizens and civil servants interact with AI agents.
When will the Deutschland-Stack 2.0 be deployed?
The timeline has three key dates: standards and governance finalized by March 31, 2026; concrete service offerings for federal, state, and municipal governments by 2028; and a legal right to digital administrative services for all German citizens by 2029.
How does the Deutschland-Stack relate to the EU AI Act?
The Deutschland-Stack’s AI layer must comply with the EU AI Act, which Germany implements through the KI-MIG (approved February 2026). The stack’s AG-UI protocol enforces human oversight requirements, its DevSecOps-by-default principle addresses AI Act risk management, and the Bundesnetzagentur serves as both the AI regulator and a stakeholder in the stack’s governance framework.
Are AI agents already used in German government?
Yes. Hamburg uses agentic AI for hydrogen infrastructure permit approvals, reducing month-long review processes to days. North Rhine-Westphalia uses AI agents for emission permit processing under the Federal Emission Control Act. The Federal Road Authority applies similar systems to infrastructure planning. The Hamburg platform won the Best Use of AI in Government Services prize at the 2026 World Government Summit.