Safe Software Bridges the AI Gap with Model Context Protocol
By Garrett Cronin, Commercial Director IMGS & DataBuilders
Safe Software has announced that Model Context Protocol (MCP) capabilities are coming soon to the FME Platform, marking a significant step in their “All-Data, Any-AI” mission. This integration is designed to help organisations move AI from experimentation into production by solving challenges related to context management, interoperability, and security. By adopting this standard, FME will act as a governed connection layer between enterprise data and the rapidly evolving world of artificial intelligence.
What is MCP?
The Model Context Protocol (MCP) is an open-source standard created to connect AI applications to external systems. It is often described as the “USB-C port for AI applications”. Just as USB-C provides a universal way to connect physical devices, MCP provides a standardised way for AI models, such as Claude or ChatGPT to interact with databases, local files, search engines, and complex workflows.
Within the FME ecosystem, MCP serves as the bridge between established enterprise systems and AI, allowing users to securely connect any AI to all their data without code and without vendor lock-in.
Why is it important?
The integration of MCP into FME is critical for several reasons:
- Freedom of Choice: Organisations are no longer forced to adapt to a single AI provider. FME Flow‘s MCP server allows you to bring your data and workflows to any model your organisations chooses.
- Standardisation Over Customisation: Traditional integrations are often hard-coded to specific models, making them difficult to scale. MCP standardises this layer, allowing you to reuse capabilities across different models as technology evolves.
- Trusted by Design: FME provides a single, governed integration layer, ensuring AI agents only connect to tools and data sources that you explicitly control.
- Eliminating Hallucinations: By allowing AI models to interact directly with real-time API’s and services rather than relying on stale training data, FME ensures AI actions remain predictable and grounded in your “source of truth”.
How it Works:
Safe Software is implementing MCP through two primary mechanisms that allow for both the consumption and provision of AI tools.
The MCPCaller Transformer
The MCPCaller Transformer is the bridge that allows FME workflows to consume tools from any MCP server, instantly extending FME’s reach to any system that supports the protocol. It features a dual-mode architecture:
- Design-Time (Deterministic): The author specifies exactly which MCP tool to use, ensuring precise execution for known workflows.
- Run-Time (Dynamic): The FME workspace evaluates intent at run time and autonomously selects the most appropriate tool from the available inventory, enabling context-aware execution.
To manage the unpredictability of external environments, FME includes Adaptive Error Handling, allowing authors to define “safety-net” pathways if a tool fails.
FME Flow MCP Server
FME Flow is being extended with MCP Server capabilities, allowing you to turn existing FME logic into secure, AI-ready tools.
- Easy Tool Creation: You can quickly transform established workspaces into tools that AI can discover and execute.
- Universal Discovery: The registry automatically generates and maintains metadata so any MCP client can “understand” how to interact with your FME workspaces.
- Enterprise Security: Every AI-triggered tool call is governed and protected by enterprise-grade controls, including OAuth 2.0 support.
What does this mean for FME Customers?
For existing FME users, this announcement represents a future-proof way to integrate AI into their operations without rebuilding integrations as new models emerge.
Real-World Impact Across Industries
- Public Sector: Agencies can use the MCP Server to securely expose approved government systems and policy data, providing 24/7 citizen support without exposing sensitive data.
- Utilities: Teams can use the MCPCaller to retrieve and validate asset registry, work order, and IoT data in real time, reducing manual lookups during outages.
- Transportation: Agencies can expose construction schedules and asset inventories as MCP tools, providing a consistent and authoritative operational view across fragmented systems.
Cost and Availability
- Safe Software has confirmed that these capabilities will be included at no additional cost for customers with FME Form and FME Flow.
The rollout will happen in phases:
- FME 2026.1: The MCPCaller Transformer will be available as a full feature.
- FME 2026.2: The FME Flow MCP Server will be introduced as a beta capability, with general availability to follow in a future release.
