sct
A fast, local-first SNOMED CT toolkit written in Rust. Convert a SNOMED CT RF2 release into queryable formats in seconds. Almost ridiculously fast on modern hardware. Free and open source. No Java. No Docker. No terminology server.
cargo install sct-rs
sct ndjson --rf2 ~/path-to-your-SNOMED-RF2.zip/
sct sqlite --input snomed.ndjson
sct lexical "heart attack"
Full walkthrough · Why build this? · Benchmarks
-
Build the pipeline
Convert an RF2 snapshot into SQLite, Parquet, Markdown, or Arrow embeddings in a single command. 831k concepts in under 30 seconds on a laptop.
-
Search
Full-text search via FTS5 for keywords and phrases. Semantic vector search via local Ollama embeddings for meaning-based queries. Both work entirely offline.
-
Connect to AI
A local MCP server exposes SNOMED CT as tools for Claude, Cursor, and any other MCP-compatible client. Ask questions about concepts, hierarchies, and relationships directly in your AI assistant.
-
Explore
A keyboard-driven terminal UI and a local web GUI for browsing concepts, navigating hierarchies, and inspecting relationships — no browser extension or remote service needed.