July was about search, and about opening up the vocabularies to people who haven't signed up yet. If you search OMOP concepts through the API, the results you get back this month are meaningfully better than the ones you got last month.

Search now finds what you actually searched for

Search has been rebuilt around a simple idea: the thing you typed should come first.

Exact name matches always surface. If a concept is named exactly what you searched for, it's at the top. It can't get lost behind thousands of partial matches any more.
Shorter, more general names rank higher. Searching "pain" now gives you "Pain" before "Post-thoracotomy pain syndrome".
The concept name counts for more than the code. A match on the name is what you usually meant. A match on a code is a fallback.
Standard SNOMED concepts are preferred when several concepts share a name, so you land on the one you'd want to store in an OMOP CDM.

And it's still fast.

Autocomplete returns real concepts, not just text

Autocomplete used to hand back a list of strings. Useful for a dropdown, useless for anything else, because you still had to run a second search to find out what any of them meant.

Each suggestion now arrives complete: concept ID, concept code, full concept name, vocabulary, domain, and whether it's a standard concept. Enough to render a proper result card, and enough to link straight through to the concept without a follow-up call.

Semantic search holds up under filters

Semantic search lets you describe what you're looking for in plain English instead of guessing at terminology. It worked well, until you narrowed it. We improved the search to handle filters better and faster.

Hierarchy results now tell you when they're incomplete

If you build cohorts from concept hierarchies, this one matters.

Cut-off lists are now labelled. When a concept has more ancestors or descendants than a single response can carry, the response says so. Previously a truncated list looked identical to a complete one, which meant a cohort could be quietly undercounted with nothing to indicate it.
Ancestors and descendants return the full set. Paging through /ancestors or /descendants now walks the entire hierarchy, and the total count reflects the whole thing rather than the first page's worth.

We've also published a new guide on building custom concept groupings using these endpoints.

Free concept pages: browse any OMOP concept, no API key

Every concept in the OMOP vocabularies now has its own public page at omophub.com/concepts/{concept_id}. All of them, roughly 11 million, no account and no API key.

Each page gives you:

The concept name, code, vocabulary, domain and concept class, plus whether it's standard
Synonyms, including non-English ones where they exist
Where the concept sits in the hierarchy, with parents and children you can click through
Mappings to other vocabularies, so you can see how a SNOMED code lines up with ICD-10-CM, RxNorm or LOINC
A ready-to-copy API request for pulling the same data programmatically

They're built to be linked. Drop a concept URL into a protocol, a data dictionary, a GitHub issue or a paper, and whoever opens it gets the full picture without needing OMOPHub access.

Try it with Type 2 diabetes mellitus.

MCP Server v1.5.3: steadier for AI agents

If you use OMOPHub with Claude or another AI assistant through our MCP server, sessions should now stay up. The issues with connection cleanup and client disconnections are fixed.

Running the MCP server yourself? Update with:

npm install -g @omophub/[email protected]

Links

Noticed a search result that still looks wrong? Tell us the query, we'd genuinely like to see it.