The latest version of FME doubles down on AI integration. MCP support now extends across both FME Form and FME Flow — but that’s not the whole story. FME 2026.3 also brings multi-geometry writes for PostGIS and Oracle, a smarter FeatureWriter that keeps running even when individual records fail, new formats, and a range of FME Flow improvements that make everyday administration less of a headache.
Read on to see how the new features will change the way you work with data.
Contents
- MCP Resources in FME Flow – Share Workspace Outputs with AI Agents
- AI Assist as an Independently Updated Plugin – No Need to Wait for the Next FME Release
- Rejected Port on the FeatureWriter – Keep Processing Even When Individual Writes Fail
- Apache Iceberg Reader – Pull Lakehouse Data Directly into FME
- Multiple Geometry Tables for PostGIS and Oracle – Skip the Manual Table Setup
- Security Settings Sync Between Form and Flow – Catch Publishing Problems While You’re Still Authoring
- Images in Workspace Annotations – Document Workflows Visually, Not Just in Text
- MCP Tools for Job Monitoring – Ask About Failures Instead of Digging Through Logs (Technical Preview)
- Better Search, Sorting, and Filtering in FME Flow – Find What You Need Without Scrolling
- OAuth for Outbound Email – Keep FME Flow’s Email Working When Basic Auth Is Blocked
- ListSorter with Multiple Sort Keys – Sort Complex Lists in One Step
- AttributeRenamer with Dynamic Naming – Rename Attributes Based on Other Attribute Values
- New Integrations in FME 2026.3
MCP Resources in FME Flow – Share Workspace Outputs with AI Agents
FME Flow now natively supports MCP Resources. MCP – Model Context Protocol – is an open communication standard that defines how AI tools (assistants, agents, chatbots) connect to external systems and data. Think of it as a shared interface: instead of building a separate integration for every tool, one common protocol handles it all.
Until now, getting FME outputs into an AI tool meant manual steps or custom workarounds. With MCP support, it happens through a standard protocol – no extra work on your end.
With MCP Resources, FME workspaces can automatically share their outputs – text files, Shapefiles, CSVs – with AI agents through standard Resource Links.
It’s important to note: an AI agent only gets access to the processed output and user parameters. It has no visibility into the workspace itself, cannot modify it, and cannot reach the source data. FME processes the data as it always has – the agent simply receives the finished result.
Want to learn more? Watch the short video: FME 2026.3 Feature Demos: MCP Resources
You’ll find more videos on FME and its individual features on the official FME YouTube channel.
AI Assist as an Independently Updated Plugin – No Need to Wait for the Next FME Release
AI Assist is now available as a standalone plugin through FME Hub. You decide when to install updates – independently of the FME Form release schedule. From the menu in Workbench, you can browse version history and changelogs and turn on automatic updates. You can also adjust settings to narrow or expand its permissions or switch it off entirely. Using the AI assistant requires signing in to FME Community.
Rejected Port on the FeatureWriter – Keep Processing Even When Individual Writes Fail
The FeatureWriter transformer now includes an optional Rejected output port. When a write fails – due to a temporary HTTP error, a constraint violation, or anything else – the affected features route to the Rejected port instead of stopping the entire process.
Initial support covers Esri Shapefile, CSV2, and PointCloud XYZ, with more formats on the way in future releases.
You can build the retry and error handling logic directly inside the workspace, without reworking your entire error management approach from the ground up.
Apache Iceberg Reader – Pull Lakehouse Data Directly into FME
FME can now read Apache Iceberg tables – the open table format at the heart of modern data lakehouses. This first release supports tables stored in Amazon S3, Azure Blob Storage, and Google Cloud Storage, as well as reading from the AWS Glue catalog.
If your organization runs a lakehouse, FME can now read from it directly – no additional middleware required.
An Apache Iceberg writer is planned for a future release.
Multiple Geometry Tables for PostGIS and Oracle – Skip the Manual Table Setup
FME can now create and write database tables with multiple spatial columns using the Geometry Definition Table, for both PostGIS and Oracle.
Previously, multi-geometry tables had to be created in pgAdmin or another SQL tool before FME could enter data into them. That step is no longer necessary, which noticeably speeds up project work.
Security Settings Sync Between Form and Flow – Catch Publishing Problems While You’re Still Authoring
A workspace that runs fine on a local machine can fail after being published to FME Flow, because Flow typically enforces a stricter security tier. FME 2026.3 addresses this with a per-user setting that makes FME Form – where you build your workspace locally – apply whichever security tier is stricter (local or FME Flow) while you’re still in Workbench. That way, you catch problems before they show up in FME Flow, not after.
No admin rights are needed. No restart required. Form’s Security page has also been redesigned with status indicators and warning banners that flag meaningful mismatches – so issues surface before you publish.
Want to learn more? Watch the short demo: FME 2026.3 Feature Demos: Security Hardening
Images in Workspace Annotations – Document Workflows Visually, Not Just in Text
Annotations in FME Workbench now support embedded images alongside text. You can upload image files directly into an annotation to include diagrams or screenshots in your workspace documentation, rather than relying on written descriptions alone.
Images are embedded in the .fmw file itself, so file size grows with every image you add — worth keeping in mind for workspaces shared across teams.
Want to learn more? Watch the short demo: FME 2026.3 Feature Demos: Annotations & Images
MCP Tools for Job Monitoring – Ask About Failures Instead of Digging Through Logs (Technical Preview)
FME Flow 2026.3 includes a built-in MCP server that exposes job history, performance trends, and error logs to any MCP-compatible AI client — including Claude Desktop. Instead of working through multiple dashboards and raw log files, administrators can ask plain-language questions to pinpoint why a job failed.
The server is off by default, authenticates via OAuth 2.0, and is restricted to the roles and users an administrator explicitly permits.
Want to learn more? Watch the short demo: FME 2026.3 Feature Demos: MCP Observability Tools (Technical Preview)
Better Search, Sorting, and Filtering in FME Flow – Find What You Need Without Scrolling
The table improvements introduced in FME 2026.2 now extend to four additional areas: User Management, Flow Apps, the Deployment Parameter Store, and Broadcast Messages.
Users and roles are searchable by name. Flow Apps tabs get owner and status filters. Broadcast Messages adds search by name plus filters for status, severity, and type. The user filter on Jobs and Schedules now accepts multiple users at once.
OAuth for Outbound Email – Keep FME Flow’s Email Working When Basic Auth Is Blocked
FME Flow now authenticates outbound email using OAuth 2.0 with PKCE, supporting both Microsoft Exchange Online and Google Workspace. The change covers everywhere Flow sends email:
- email automation actions,
- notification subscriptions,
- and system emails including password resets.
Organizations that enforce MFA or have disabled basic authentication can keep Flow’s email working without any additional configuration.
ListSorter with Multiple Sort Keys – Sort Complex Lists in One Step
The ListSorter transformer now supports multiple sort criteria — bringing it in line with the capability the standard Sorter has had for some time. Previously, chaining ListSorters in order to sort by more than one criterion didn’t reliably preserve order when values were duplicated. Now you can define all sorting priorities in a single transformer.
AttributeRenamer with Dynamic Naming – Rename Attributes Based on Other Attribute Values
The AttributeRenamer now accepts Value (AttributeName) as the target name for a renamed attribute. Attribute names can be assigned dynamically based on the content of other attributes – no extra scripting or custom transformers are required.
New Integrations in FME 2026.3
OpenDocument Spreadsheet (.ods) reader — Reads the ISO-standard spreadsheet format used in government and open-standards environments. Writer support is planned for a future release.
OpenDocument Text (.odt) writer – Produces ODF-compliant word-processing documents from FME workflows, building on the .odt reader released in FME 2026.2.
Mappedin MVF reader and writer – Reads and writes Mappedin’s Map Vector Format for indoor mapping. The format covers multi-floor venues, individual floors, and spaces, enabling integration of indoor spatial data with other spatial datasets.
PLY (Polygon File Format) reader – Reads 3D meshes and point clouds from photogrammetry and laser scanning tools. Writer support is coming in 2026.4.
CIM XML reader and writer – Exchanges Common Information Model data between power utility systems using the IEC standard format, directly within FME workflows.
Cesium 3D Tiles writer – v1.1 – Writes 3D Tiles 1.1 with full attribute support and GLTF output. Handles larger datasets, improves rendering in web clients, and removes the need to manually pre-tile data before publishing.
WebODMConnector – Generates raster orthomosaics from drone and aerial imagery directly inside FME, eliminating a step that previously required a separate third-party tool.
Want to See How FME 2026.3 Works for You?
FME 2026.3 is available now. If you’d like to take a closer look at a specific feature or find out whether the latest version of the Platform is a fit for your environment, don’t hesitate to contact us.
This article is based, in part, on material by Safe Software: FME 2026.3 – Feature Highlights | Community
