Azure Logic Apps in 2026: agent loops, MCP and a new SaaS edition
Microsoft has turned Logic Apps into a platform for AI agents, with agent loop, workflows as MCP tools and the new Logic Apps Automation. An overview for Azure users.
Azure Logic Apps has long been the workhorse in Azure Integration Services: workflows with connectors to hundreds of systems, often alongside API Management and Service Bus. Over the past year, Microsoft has turned Logic Apps into something more: a platform where AI agents and fixed business logic run side by side. For organisations already on Azure, that is worth understanding before buying a separate agent platform.
Agent loop: agents inside the workflow
Agent loop reached general availability in November 2025. The idea is simple: a step in a Logic Apps workflow can be an AI agent that gets a goal, a model and a set of tools. The tools are ordinary Logic Apps actions, so the agent can use the same connectors and the same managed identities as the rest of the workflow. The workflow decides where the agent is allowed to act, and deterministic steps can check the result before anything is written back.
Microsoft describes this as agentic workflows, where AI agents and deterministic logic work together in orchestrated systems. That is a sober approach: the agent does the parts that need judgement, and the workflow does the rest.
Workflows as MCP tools
At Build 2026, the Logic Apps MCP Server reached general availability. Existing workflows can now be exposed as tools that agents can discover and call over the Model Context Protocol. For an organisation with hundreds of Logic Apps workflows, this means that years of integration work can be reused by agents in Copilot, Foundry or other MCP clients, without rebuilding anything.
Logic Apps Automation: a new SaaS edition
Microsoft also announced Azure Logic Apps Automation in public preview at Build 2026. It is a managed SaaS edition that packages workflows, AI agents, knowledge services and model access in one environment, with isolated compute per project, built-in identity, RBAC, audit logging, VNET integration and private endpoints.
It includes Knowledge as a Service, a fully managed RAG pipeline, and supports three ways of working with agents: agent loop, calling Microsoft Foundry agents directly from workflows, and a managed sandbox for running tools such as GitHub Copilot in isolation.
Microsoft positions it between Power Automate, for personal productivity, and Logic Apps Standard, for complex enterprise integration. The target is teams that want enterprise security without having their own integration developers.
In addition, Codeful Workflows lets .NET developers build workflows code-first with the Logic Apps Standard SDK.
How it compares
Microsoft was named a Leader in Gartner’s Magic Quadrant for iPaaS for the eighth year in a row in 2026. The strength is the depth of the Azure platform: identity through Entra ID, networking, API Management with AI gateway capabilities, and Foundry for models and agents. The weakness for some organisations is the same: the value is highest if you are already committed to Azure.
The other iPaaS vendors are moving in the same direction, with agents and MCP support on their roadmaps. The difference is that the independent vendors promise to be neutral across clouds and AI vendors.
What Azure users should do
- Map existing workflows that would be useful as tools for agents, and test the MCP Server on a few of them.
- Start agent loop in a bounded process where a person can approve the result, such as case handling or classification.
- Look at Logic Apps Automation if business teams are building automations in Power Automate that have outgrown it, but remember that it is still in preview.
- Put API Management in front of MCP servers and model traffic, so that access, logging and cost are governed in one place.
Sources
This post was written with AI assistance and reviewed by the editor before publishing.