ConferenceGrid

npx add-mcp 'https://conferencegrid.com/api/mcp'

MCP server

Ask Claude or ChatGPT about conferences and get answers from a catalogue that re-verifies itself daily, rather than from whatever the model remembers. Read-only. Sign in once with your free ConferenceGrid account; there is no API key to manage.

Endpoint

https://conferencegrid.com/api/mcp

Use the address exactly as written. The www form redirects, and most clients turn that redirect into a request this endpoint refuses, which looks like a broken server rather than a wrong address.

Connect it

Every client follows the same three steps: paste the address, sign in to ConferenceGrid in the browser window that opens, click Allow. Sign-in is a one-time link to your email. If you have no account yet, one is created for you at that moment. Nothing to copy, no key to keep.

Claude — Settings, then Connectors, then Add custom connector. Paste the address and leave the OAuth fields empty; Claude discovers them. Click Connect and finish the sign-in.

Claude Code — claude mcp add --transport http conferencegrid https://conferencegrid.com/api/mcp then /mcp, pick conferencegrid, and choose Authenticate.

ChatGPT — add it as a custom connector with the same address and complete the sign-in when asked. The server carries OpenAI's search and fetch contract alongside the tools below, so one address serves both clients.

Anything else that speaks Streamable HTTP and OAuth 2.1 with PKCE works too. Discovery is standard: the endpoint answers an unauthenticated call with a 401 that names its resource metadata, and the authorization server metadata sits at /.well-known/oauth-authorization-server. Clients may identify with a client metadata document URL or register dynamically. The server is stateless, so there is no session to hold open.

Connected before 19 September 2026? The server was anonymous until then. Remove the old connector and add it again, or in Claude Code run /mcp and authenticate.

What to ask

Four prompts ship with the server and show up in the client as conversation starters. Or ask in your own words:

• Which conferences does Datadog sponsor in the next six months?
• Find fintech conferences in Berlin before June, and tell me which have a sponsor prospectus.
• What changed on KubeCon Europe since last month?
• Which security conferences are still open for speaker submissions?

The tools

Nine, all read-only. Nothing here can write, delete or spend money.

search_conferencesUpcoming events by city, country, topic or date window.

get_conferenceOne event with its speakers, sponsors, exhibitors and sessions.

company_eventsEvery event a company appears at, as sponsor, exhibitor or speaker employer.

events_open_for_sponsorshipWho is taking sponsors now, plus the comparable events a company has missed.

find_calls_for_speakersOpen calls, with deadlines and submission links.

organizer_contactHow to reach an organiser, as a link to the event's page.

whats_newWhat changed for a company or a conference since a date.

search / fetchThe ChatGPT connector contract, over the same data.

Before you trust an answer

Every record carries the date we last verified it. That is the whole point. A number with a date on it is a different claim from a number without one, so ask for it if the model does not offer it.

An empty list is an answer, not a failure. Roughly one upcoming event in five publishes before its line-up does. No sponsors listed means we looked and the organiser has not published them yet. The tools return an explicit empty list so the model can tell that apart from not having looked.

Coverage thins further out, and that is seasonal. We hold about a year per conference, and organisers publish next year's dates after this year's event runs. A quiet month eighteen months out is normal. Compare a month to the same month a year earlier, never to the month beside it.

Company coverage depends on spelling. A company that writes its own name three ways across three event sites may not have collapsed into one record yet. Treat a company's event footprint as strong evidence, not a complete census.

Privacy and safety

The server knows which account is calling, because you signed in. That is all it knows: the client app never sees your email address, and the tools return no user data of any kind. We record which tool was called, by which account and from which client app, how many rows came back and how long it took, so we can see what people actually need. We record the shape of a question, never its words.

Access is yours to withdraw. Disconnect the connector in your client and its token stops working within the hour. Tokens are bound to this endpoint alone; one issued here is useless anywhere else, and one issued elsewhere is refused here.

Organiser email addresses never leave the server. The organiser tool returns a link to the event's page, and it declines events less than 21 days away, which is too late for an organiser to act on an approach anyway.

Scraped text is stripped before it reaches your model. Every description and biography here came from someone else's website. All of it goes through a filter that removes markup and instruction-shaped text on the way out, because a server that pipes raw scraped prose into your context is an injection path. Where you see (removed) in a description, that is this filter.

Limits

• Sixty calls a minute per account.
• Every tool caps its rows and states the cap in its own description. Nothing can walk the whole catalogue.
• whats_new reports records that appeared and events that were edited. It cannot yet report an edited record, so a sponsor moving from silver to gold does not show up.

Something wrong?

The data is read from conference websites, so some of it will be wrong. Tell us through the contact form with the event name, or read how the catalogue is built first. A wrong record reported is more use to us than a polite silence.

Last reviewed 2026-09-19.