TL;DR: IDC MCP server and REST API v3 are live!
Note: both are in beta - API can change as we continue testing, collecting user feedback and improving the capabilities. Please respond to this thread with any suggestions!
Natural language interface to IDC is improving!
Beta release of IDC MCP server is public and ready to serve your needs. MCP stands for Model Context Protocol - a rapidly evolving open standard interface allowing your LLM assistant to connect to specialized tools and databases. MCP is broadly supported across AI vendors, is easier to set up, and in many cases will be much faster as compared to IDC skill, and much more accurate as compared to vanilla LLM. To leverage IDC MCP interface, all you need to know is this MCP server URL: https://api.imaging.datacommons.cancer.gov/mcp. Just ask your LLM assistant what are the setup instructions specific to the product you are using!
Read more about IDC integration with AI assistants in this new documentation section: https://learn.canceridc.dev/ai-assistants/agents.
MCP endpoint: https://api.imaging.datacommons.cancer.gov/mcp.
Note that the latest version of IDC agent skill is MCP-aware, and will use MCP as preferred tool if it detects it is installed in your environment.
Once IDC MCP and/or skill is set up, requests like these should work without further prompting:
- Find breast MRI in IDC, show the counts and total size, and give me a download command.
- Which IDC collections have prostate MRI with expert segmentations?
- How much CT data is there for lung cancer, and how much of it is cleared for commercial use?
- Generate citations for the collections in my cohort.
IDC REST API has been overhauled and v3 is live!
Some of you noticed that IDC REST API v2 was … really slow. The reason was the design choices we made at the time that required a call to IDC Portal and a BigQuery query for typical calls. Enter v3, with its implementation completely self-contained with all requests resolved directly on the server via idc-index - the same Python library that powers IDC MCP server.
With v3 we not only redesigned the API to make it more efficient, but also aligned it with the idc-index Python API that proved useful over the past years. Plus MCP and REST surfaces have been developed in coordination to simplify implementation and maintenance.
To get started with IDC REST API v3 check out this documentation page: https://learn.canceridc.dev/rest-api/api.
API endpoint: https://api.imaging.datacommons.cancer.gov/v3