Agent-Enabled OHIF Viewer

• 

  The medical image viewer in action

Inspiration

I have a family member who is a radiologist and uses OHIF's medical imaging viewer. After taking a look at his workflow it was clear that he spends a lot of time arranging and displaying medical images; once he finds an axial slice from a CT scan, he'll usually want to view it from different anatomical planes (axial, coronal, sagittal). Since setting this up is tedious, I wanted to explore how web-mcp could help automate this process.

What it does

I've added 14 web-mcp tools for searching and opening studies, configuring viewports, navigating image stacks, adjusting visualization, and managing measurements.

How I built it

I spoke with a family member to confirm which parts of his OHIF Viewer workflow were the most tedious. I then had Codex explore the codebase to identify the business logic I could use to connect the web-mcp tools to.

Challenges I ran into

The implementation wasn't too difficult but it was important for me to review the code to check for edge-cases, which revealed I was exposing patient names and birth dates to the model, which I then cleaned up and redacted.

Accomplishments that I'm proud of

I haven't had time to perform a full eval, but the time it takes to open a CT scan to then render a complete hanging of the anatomical planes is dramatically reduced while (presumably) reducing context bloat.

What I learned

High-stakes workflows benefit from a clear division of labor: agents handle navigation and preparation using web-mcp tools while medical professionals keep their judgment and final approval. I could've added tools to run analyses but I don't think web-mcp is a great fit here.

What's next for Agent-Enabled OHIF Viewer

I plan on opening a PR in the OHIF Viewer repository with the changes made here, as I'm hoping we can get this added to the list of official OHIF Viewer extensions.

Built With

• docker
• react
• typescript