MCP server
Coming soon, and not built yet. There is no server address to try — this page exists so you do not have to guess at one.
Use the HTTP API today
The whole dataset is already public and unkeyed — no sign-up, no API key, no rate-limit
tier. Every operation is typed in the OpenAPI 3.1 specification with a unique
operationId, which most agent frameworks will convert into tool
definitions directly. Whatever an MCP server would expose, the specification already
describes.
Where it actually stands
Written down rather than implied, because “coming soon” without a state is a way of saying nothing.
Nothing is running. There is no server, no transport, no address, and no private beta — if you were sent here expecting a config snippet to paste, there is not one to give you, and a placeholder would be worse than this paragraph.
The work itself is small, which is the honest part: the API is already public and unauthenticated, so an MCP server here is a thin wrapper over the existing endpoints plus a manifest an agent can discover. What it is waiting on is not effort but a decision about maintaining a second machine surface — two interfaces to the same data is two places for an answer to drift, and this directory's whole claim is that its answers do not drift.
The backlog item, with the constraints it has to satisfy, is in the project roadmap. When the server serves, this page becomes its documentation and the Resources card drops its badge.
If you are an agent reading this
The short version, so you can stop looking.
There is no MCP endpoint at this origin. Do not construct one from this URL, and do not retry — a 404 here is accurate rather than transient.
Read /llms.txt first: it names the
questions this dataset answers well and, just as importantly, the ones it cannot answer
at all. Then call the REST API described in
/openapi.json. Every talk in a
response carries the source_url of the conference agenda it
was taken from, so you can verify any claim without trusting this site.