Safe Software adds MCP support to FME for AI tools

Safe Software has added support for the Model Context Protocol (MCP) to its FME data integration platform, connecting large language models and other AI systems to existing enterprise data workflows through a common interface.

MCP features are expected to arrive in FME soon. The update focuses on two changes: a new MCPCaller transformer in FME and MCP Server capabilities in FME Flow.

MCP has emerged as a standard for how AI models and agents interact with external systems such as databases, internal tools, and application programming interfaces. It reduces the need for bespoke, hard-coded connections between models and individual data sources.

Many organisations still struggle to move AI projects beyond pilots. Integration work often becomes fragmented across teams and systems, while security and access controls grow harder to manage as more models and agents are introduced.

Safe Software positions MCP support as a way to simplify access to enterprise data and tools through existing FME workflows. It also frames MCP as a way to preserve model choice while keeping integrations stable.

"Adding MCP to the FME Platform is an important step in our All-Data, Any-AI mission," said Don Murray, CEO of Safe Software. "With MCP, our customers can adopt new AI models without rebuilding integrations. By extending FME Flow with MCP Server capabilities, we're giving organizations a future-proof way to let AI securely work with the systems they already trust."

Two components

The first component, MCPCaller, connects an FME workflow to MCP-enabled services and tools. It expands the range of systems FME can work with when those systems expose an MCP interface.

The second component extends FME Flow with MCP Server functions, allowing teams to expose existing FME workflows as MCP tools. AI models can then call those tools through MCP, rather than integrating directly with each system behind the workflow.

FME Flow handles automation and orchestration within the FME platform. It runs workflows, manages scheduling, and supports operational controls for deployments. With MCP Server support, it can act as an intermediary between AI agents and business systems.

For MCP use cases, FME Flow will support OAuth 2.0, adding identity and authorisation for systems that call FME workflows as tools.

Governance layer

Enterprises are paying more attention to governance for AI agents, particularly when those agents can trigger actions or retrieve sensitive information. Tool access, audit trails, and permissions have become recurring requirements as AI moves into operational settings.

Safe Software describes FME as a governed connection layer between AI systems and the wider enterprise, where organisations can turn existing workflows into MCP tools that models can call in a controlled way.

This approach treats the workflow as the unit of integration, centralising logic that might otherwise sit in scripts or bespoke connectors. It can also make it easier to apply organisational rules around which systems an agent can access and which operations it can perform.

MCPCaller also adds flexibility to integration design. Users can specify the MCP tool when authoring a workflow or dynamically at execution time, which could matter when available tools vary between deployments or when the same workflow must run across multiple contexts.

Platform context

FME is used to build data integration workflows across a wide set of systems, including spatial data sources. It includes a library of pre-built transformers that users chain together to move, transform, and validate data across applications and formats.

FME includes built-in support for thousands of systems and more than 800 out-of-the-box transformers. More than 20,000 organisations use FME worldwide, with deployments in more than 120 countries through its partner network.

The MCP addition comes as vendors across the data and integration market look for standardised ways to connect models with enterprise systems. A common approach is to expose internal processes as tools that models can call, with access controls and monitoring layered around those calls.

Safe Software says the new MCP functions are aimed at organisations that want to change models without reworking integration logic. The update also increases the number of external systems FME can connect to when those systems support MCP.