NA-MIC Project Weeks

Back to Projects List

Redesigned REST API and MCP server for Imaging Data Commons

Key Investigators

• Andrey Fedorov (BWH, USA)
• Bill Clifford (Institute for Systems Biology, USA)
• Steve Pieper (Isomics Inc., USA)
• Martin Bellehumeur (Radical Imaging, Germany)
• Ryan Cahill (, USA)
• Ulrike Wagner (Leidos Biomedical Research, USA)
• Granger Sutton (National Cancer Institute, USA)

Project Description

Questions? Want to participate remotely? Join our project Discord channel!

TL;DR: Work on v3 of Imaging Data Commons (IDC) REST API simultaneously with the development of the IDC Model Context Protocol server.

IDC already offers a customized agent skill in https://github.com/ImagingDataCommons/imaging-data-commons-skill, which is very powerful in helping with the use of IDC. However, in order to use it effectively, the user needs to have access to an LLM platform that supports code execution, and needs to suffer the “cold start” cost of setting up the environment. IDC MCP interface can provide “zero startup cost” solution for answering at least some of the queries, optimized for the use by agents. At the same time, MCP interface will need to rely on IDC API, which in its current v2 form is not useful at all. We are in the process of simultaneous redesign of IDC API and design of the MCP interface. Since we do not have expertise on the team doing something like this before, we would appreciate community feedback.

Objective

1. Maximize utility of the revised IDC API and new MCP server to IDC users, identify design flaws before public release.

Approach and Plan

1. Refine and deploy the initial implementation.
2. Provide test URL endpoints for interested users.
3. Test against use cases raised in the community.
4. Collect feedback from IDC users and those who have more expertise designing REST API and MCP interfaces.

Progress and Next Steps

1. Developed v3 of IDC REST API and the initial version of MCP server - repo and overview
2. Shared MCP server endpoint with interested users, collected feedback (in particular, identified the need to improve support of clinical data)
3. Improved MCP integration with Martin’s project https://github.com/mbellehumeur/vtk-js/pull/1
4. Implemented support for clinical tables query.
   • public linkt to a conversation with Claude demonstrating the capabilities of clinical data exploration with the updated IDC MCP server
5. Summarized (but not benchmarked!) differences between using “vanilla” Claude vs Claude + IDC skill vs Claude + IDC MCP.
6. 3rd party security review by Ryan completed.

Illustrations

Snippets from this conversation

Background and References

No response