▸ saaro:~/blog ▊ Notizen & Artikel DE
← back

📷 "BMW 5 series interior." by Rob Ellis' is licensed under CC BY 2.0. To view a copy of this license, visit https://creativecommons.org/licenses/by/2.0/.

Backstage 2026: MCP, AI Agents, and the Future of Developer Portals

17 August 2026 · 4 min · Martin Jochum #Backstage#Developer Portal#Platform Engineering#MCP#KI#DevOps#CNCF

Backstage turns five, and the developer portal from Spotify is in the midst of perhaps the most important transformation in its history. After Backstage released the New Frontend System for production in 2025 and celebrated its fifth birthday, 2026 is all about AI integration. With the introduction of the Actions Registry, MCP server support, and a growing number of AI-powered plugins, the question arises: Will Backstage become the nervous system of AI-driven software development — or is it, as critics say, a fading model?

MCP and Actions Registry: Backstage becomes an AI access point

The most significant innovation of 2026 is the official MCP Actions Backend for Backstage. The Model Context Protocol (MCP) is an open standard that allows AI assistants and LLM clients to access external tools and data — essentially a USB-C port for AI. Backstage now registers its actions (e.g., “Create service from template”, “Search catalog entry”) in the Actions Registry and provides them as MCP tools. AI agents in VS Code, Cursor, or other MCP-capable environments can thus directly access the software catalog, execute scaffolder templates, and search catalog entries without a human developer needing to open the Backstage interface.

The official Backstage blog describes this development as “AI Superpowers”: The combination of Actions Registry and MCP server makes Backstage a central interface between humans and AI agents. In the future, a developer can simply tell their AI assistant: “Create a new Python microservice with PostgreSQL connection and register it in the catalog” — the assistant then uses Backstage as an orchestration layer without the developer having to execute the individual steps manually.

In parallel, community projects are emerging, such as the Standalone MCP Server for Backstage (backstage-mcp), which works without plugin installation and functions as a slim 500 KB bridge between LLM clients and the Backstage REST API.

AI plugins and Infrastructure-as-Code from the assistant

The plugin ecosystem is growing rapidly. The RAG AI Assistant allows direct access to Backstage catalog data via LLM prompts — for example, to answer questions like: “Which microservices in my team had incidents related to deployments last week?” The Backchat GenAI Plugin also integrates private, local LLMs into the portal, so companies do not have to send sensitive data to external APIs.

Going a step further is StackGen, an AI-powered plugin that automatically generates deployment-ready Terraform configurations from Backstage service metadata. Instead of just cataloging infrastructure, StackGen creates it — tailored to the organization’s templates, policies, and naming conventions. The plugin is considered a pioneer of a new generation of “AI-powered infrastructure automation,” which Gartner predicts will handle around 15% of all infrastructure tasks semi-autonomously by 2028.

Controversy: Is Backstage a fading model?

Not all voices in 2026 are positive. Zohar Einy, CEO of Port.io, caused a stir in April with the provocative blog post “Backstage Is Dead”. His core criticism: Backstage is a React framework that forces platform engineers to master a technology unrelated to their core business (Terraform, Kubernetes, CI/CD). The costs are high: two to five full-time employees for operation, six to twelve months to a productive setup, and the true total cost is around $150,000 per 20 developers. Additionally, plugins operate in their own silo databases with no overarching data model — queries across multiple data sources are hardly possible.

The Backstage community’s response: The New Frontend System massively simplifies plugin development, and the MCP integration is precisely making Backstage the open, AI-capable layer that critics miss. More than 3,400 companies now use Backstage, and the CNCF incubator counts over 270 known adopters and around 230 plugins.

Platform economy: Numbers and outlook

The market for Platform Engineering continues to grow: from an estimated $5.76 billion (2025) to a projected $47.32 billion by 2035. According to analysts, Backstage holds a market share of around 89% in the Developer Portals segment. Toyota, for example, was able to realize total savings of around $10 million over two years through the introduction of Backstage.

Conclusion

Backstage 2026 stands at a crossroads: The MCP integration and the new frontend system are strong signals that the project wants to make the leap from a pure developer portal to an AI-powered platform orchestration layer. The criticism regarding complexity and operating costs cannot be dismissed, but the direction is right — Backstage is opening up to AI agents, standardizing its API, and lowering the entry barriers for new plugin developers. Whether this is enough to convince smaller teams as well will be seen in upcoming releases. What is certain is that Backstage remains the central reference point for Internal Developer Platforms in 2026 — and, at the latest with the MCP integration, becomes the indispensable link between humans, machines, and microservices.

Sources

🌐 Machine-translated from the German original, editorially reviewed. 🤖 Written with AI assistance.