Key Features - FME 2026.2 Release

FME 2026.2: Native Looping, MCP, and Better JSON Handling

Article written by Michael Studdert, Pre-Sales Consultant & Account Manager

Michael Studdert recently explored the latest FME 2026.2 release and its impact on enterprise integration, automation, and AI-enabled workflows. With organisations increasingly seeking ways to connect systems, streamline processes, and leverage emerging AI technologies, this release introduces several capabilities that make FME more powerful and accessible for developers and business users alike. Among the many updates, three enhancements stand out: Native Looping, FME Flow as an MCP Server, and improved JSON handling.

One of the most notable improvements is Native Looping. Looping has been possible in FME through custom transformers, but implementing it often required complex configuration, particularly when blocking or grouped processing was involved. The new approach brings looping directly into the workspace via a bookmark, making iterative processes easier to build, understand, and maintain. Rather than logic sitting inside custom transformers, authors can now see the workflow directly on the canvas, making troubleshooting and handover to other developers much simpler. By reducing the complexity associated with traditional looping methods, users are spending less time managing execution behaviour.

Another significant addition in FME Flow is the ability to publish workflows and expose data as an MCP Server. MCP (Model Context Protocol) is becoming a common way for AI assistants to discover and use external tools. MCP Tools are created to extend and integrate AI tools with external systems. An MCP Server can provide a suite of tools that provide access for the agent to access data, update and create data in the integrated systems, depending on the requirements of the integration. In FME 2026.1, the MCP functionality released was through an MCPCaller transformer, enabling FME to call out to trigger external MCP tools. With this release, FME workspaces can be exposed as MCP tools, allowing platforms such as Copilot, ChatGPT, and Claude to connect to FME Flow via the MCP Server Registry and trigger FME workflows directly.

The release also introduces major improvements to JSON handling. JSON is now treated much more like a traditional FME format, with improved support for schemas, nested objects, arrays, and flexible document structures. In addition, two new transformers have been added, making it easier to generate complex JSON documents without relying on templates, text manipulation, or custom scripting. For example, the JSONObjectBuilder provides an attribute mapping interface, complete with attribute type. Once configured, it builds JSON objects from flat features, auto-building the JSON structure.

These enhancements will be valuable for API Integration, Enterprise AI use cases, and are indicative of the all-data, any-AI direction that Safe Software are taking, releasing functions that strengthen its place in enterprise integration. Building nested request and response payloads can now be handled more naturally within FME, reducing development effort and improving maintainability.

FME 2026.2 focuses on simplifying common development patterns while expanding integration opportunities. When building iterative workflows, exploring AI-driven automation, or creating modern API integrations, these enhancements help make FME a more capable and cost-effective enterprise integration platform.