Is validating of data in the MCP writer always the right thing to do?

Hi ​@martinkoch - thanks for sharing this.

The MCP Writer currently tries to access and download the file at the provided filepath in order to host it as a Resource on the MCP Server. That works fine for files within FME Flow's scope (local or Shared Resources), but it can't reach external URLs like SharePoint — hence the error.

The MCP spec actually distinguishes between these two scenarios. For https:// URIs, the spec notes that servers should only use that scheme when the client can fetch the resource directly from the web on its own — meaning it doesn't need to go via the MCP server. Your case is exactly that: Claude (or whichever MCP client) could fetch a SharePoint link directly, so the MCP Server doesn't need to hold the file at all. The current behavior may be a bit more restrictive than is required.

The most reliable path today is to download the SharePoint file within the workspace first (for example using the SharePoint Connector), then write the local copy as the Resource Link. Not ideal, but it keeps the resource within Flow's reach.

On the topic of optional validation, this is something we're actively looking at. The direction we're considering is that when an https:// URI is provided, the MCP Writer simply passes it through as a resource link without trying to access the underlying file — leaving it to the client to resolve.

No timeline to share yet but I'll make sure it's tracked.

Always appreciate the detailed questions/thoughts!

Highlighting ​@martinkoch’s related Idea in case others want to upvote it: 

Hi ​@martinkoch - just a quick update. We’re hoping to add the functionality I described in my last post into the 2026.3 release. We’re doing some work on Resources in MCP overall for this release, and it’ll fit nicely into it.

I’ll closely follow the MCP developments in FME Flow 🤓