An AI assistant can now call the engine instead of guessing
OpenOnco ships an MCP server, so an assistant answering an oncology question can route through the rule engine and return cited output rather than improvising a regimen.
Written retrospectively and published on July 19, 2026.
Doctors already ask general-purpose AI assistants about treatment. The assistants already answer. Neither of those facts is going to change because a project charter disapproves.
What can change is where the answer comes from. OpenOnco now ships an MCP server, so an assistant that supports the protocol can call the actual rule engine and return the engine's cited output — instead of generating a plausible-sounding regimen from its own weights.
The four tools
engine_info— what the engine is, and what it explicitly is not.list_diseases— the disease catalog actually covered.generate_treatment_plan— a cited plan; auto-routes to the diagnostic path when there is no confirmed histology.generate_diagnostic_brief— workup steps when the diagnosis is not settled yet.
Also shipped
/llms.txt and /llms-full.txt are live, describing what this site is for, how to cite it, and which pages are synthetic examples rather than patient records. Page metadata was enriched at the same time. Release v0.1.3 collects the lot.
The install instructions and the "build your own" notes are in mcp_server/README.md. It is deliberately dependency-free at the bridge layer, so the engine wrapper is testable without the MCP SDK installed at all.
Comments
Threads are public and indexed — please keep personal medical data out of them. Found a clinical error? Open an issue instead.
Comments are not switched on yet — the backend still needs to be configured.