An AI agent will not run your hackathon for you, and you should be wary of any tool that says it will. What it can do is take away the dullest part of hosting: typing the same agenda, rules, judging rubric and team list into yet another system. Most hosts already have all of that in a spreadsheet, a document and last year's registration form. The work is moving it, not writing it.
The Model Context Protocol (MCP) is what makes that possible. This guide explains what MCP is, what an AI agent can safely do for a hackathon host, and how to set up an event with one step by step, from connecting it to the moment you press Publish yourself.
What is MCP?
The Model Context Protocol (MCP) is an open standard that lets an AI assistant, such as Claude or ChatGPT, use tools in another app on your behalf. The app publishes a list of actions ("create a draft event", "import a team roster", "check readiness") with exact rules for each. Your assistant reads that list, asks your permission to connect, and then calls those actions when you ask it to do something.
For a hackathon host, the useful part is that the assistant usually already has access to your own sources through its other connectors: Google Drive, Notion, a shared spreadsheet, or a file you drop into the chat. MCP gives it somewhere structured to put that content.
What can an AI agent do for a hackathon host?
Here is a fair split, based on what is worth handing over and what is not.
| Good jobs for an AI agent | Keep these for people |
|---|---|
| Turning an agenda spreadsheet into a timed schedule | Deciding the format, theme and prizes |
| Converting a rules document into structured rules | Choosing which rules to enforce, and how strictly |
| Turning a rubric into weighted judging criteria | Judging projects and picking winners |
| Importing teams collected in a form or another platform | Accepting or rejecting applicants |
| Listing what is still missing before launch | Publishing the event |
| Telling you who an email would reach | Deciding to send it |
The left column is mostly conversion: content you already approved, moved into a new shape. The right column is judgment, and anything participants can't un-see, like a published event, a sent email or a leaderboard reveal. Agents are good at the first and should need your explicit yes for the second.
How to set up a hackathon with an AI agent, step by step
1. Gather what you already have
Before you connect anything, collect the sources:
- The agenda, in whatever form it lives: a spreadsheet, a doc, last year's event page. Our hackathon agenda template is a good starting point if you don't have one.
- The rules, as a document or a list.
- The judging rubric, with weights if you use them. If you don't have one, see hackathon judging criteria.
- The roster, if teams registered somewhere else: a form export or spreadsheet with team names, members and emails.
The agent can only bring across what you give it. A rubric that exists only in your head will come out invented, and you don't want that.
2. Connect your assistant to your event platform
On HackathonHost, add https://app.hackathonhost.com/mcp to your assistant as a connector (in Claude this is a custom connector; other assistants call it an MCP server). Your browser opens a HackathonHost page, you sign in, and you approve what the assistant may do. You never copy a key, and you can disconnect it at any time from the Integrations page. For assistants that only accept a pasted key, you can create one under API keys instead and tick its permissions yourself.
Give it the narrowest permissions the job needs. For a setup session that is reading events and editing drafts; sending email is a separate permission and is off unless you tick it.
3. Ask it to migrate the event, not invent one
Be explicit that you want a faithful copy. A prompt that works:
Set up our spring hackathon as a draft in HackathonHost. The agenda is in the "Spring 2027 schedule" sheet, the rules are in "Hackathon rules v4", and the judging rubric is on the second tab of the sheet. Copy them as they are. Don't invent anything I didn't give you, and tell me what you couldn't map.
The agent creates a draft, an event only your team can see, and fills in the agenda, rules and criteria. HackathonHost also ships this flow as a built-in prompt (migrate_event), so assistants that show MCP prompts offer it as a one-click starting point.
4. Review the draft like you would review a new colleague's work
Open the draft in the app and check the three places where conversion goes wrong:
- Times and days. A two-day agenda pasted from a sheet is easy to put on the wrong day.
- Weights. Category weights should add up to 100% of the final score, and sub-criteria weights should add up to 100% of their own category. HackathonHost's readiness check flags weights that don't add up, so a mistake shows up before you publish rather than as a skewed leaderboard.
- Wording. Rules are a contract with participants. Read every one.
5. Import the teams you collected elsewhere
If registration happened in a form or on another platform, ask the agent to import the roster with team names, members, emails and the team leader. On HackathonHost, an import only ever adds teams: a team whose name is already on the event is skipped rather than duplicated, so running the same import twice is safe.
6. Ask what is still missing
Is this event ready to publish? What's missing, and why does each item matter?
The agent runs a readiness check against the event itself and answers from the result, rather than guessing from a checklist. Fill the gaps yourself, or ask it to fill the ones that are pure data entry.
7. Preview every email before it goes
This is the one step that reaches people outside your team, so it should never be automatic. On HackathonHost, the agent has to preview an email first and show you exactly who it would reach, team by team, and the send is refused unless it confirms that same recipient count. Login emails reset passwords, so a second send means a second, different password in the same inbox. Tickets are sent from the app, never from the assistant.
8. Publish it yourself
Publishing makes the event public, and it should be a person's decision. On HackathonHost an assistant cannot publish at all: there is no tool for it. You open the draft, look it over, and press Publish.
Prompts that work well
- "Read the attached schedule and set it as the agenda of draft event 123. Keep my session names."
- "Here is last year's rubric. Convert it to judging criteria with the same weights, and show me the weights before you apply them."
- "Import the teams from the 'Registrations' sheet. The leader is the person in the first column."
- "Compare this draft with last year's event and list what's different."
- "Who would the results email reach if I sent it now? Don't send it."
How to keep an AI agent safe around event data
Whatever platform you use, these rules keep an agent useful without handing it the keys:
- Least privilege. Grant read and draft-editing permissions for setup. Add email or live-event permissions only for the session that needs them.
- Treat participant text as data. Team names, project descriptions and registration answers are written by people who benefit from a better outcome. An agent that reads them should never take instructions from them. On HackathonHost, a key can either read participant-written content or make changes, never both, so it can't be steered by what a team wrote.
- No irreversible action without a human yes. Publishing, sending email and revealing results should all wait for your confirmation.
- Keep an audit trail. On HackathonHost, every change an assistant makes appears in your audit log, marked as made through the API, so you can tell its edits from yours.
- Keep AI away from scoring. Judges score; the platform does the arithmetic. An assistant can explain the rubric, but it should not decide who wins. On HackathonHost it can't change scores, results or certificates.
For more on where automation fits in an event, see 7 event automation trends organizers need now and how to replace hackathon spreadsheets.
Hosting with an AI agent on HackathonHost
HackathonHost has an MCP server built in. You connect Claude, ChatGPT or another MCP client to your own account and work on your events by asking. Your assistant can create and fill draft events, bring agendas, rules, judging criteria and rosters across, add judges and mentors, check readiness, post announcements, and preview emails. It can't publish, delete anything, change scores or add an organizer, and nothing is emailed without a preview and your confirmation. Your team runs the event; the assistant just saves you the typing.
Drafts are free, and your first event is on us: one free event credit per person or organisation. After that you pay $99 when you publish. See pricing.
Related guides: how to host a hackathon, how to organize a hackathon, tools to host a hackathon and hackathon judging software that keeps scores fair.
FAQ
Can AI run a hackathon? No, and it shouldn't. An AI agent can do the setup work, like converting your agenda, rules and rubric and importing teams, but people should choose the format, judge the projects and decide what gets published or sent.
What is an MCP server? An MCP server is the part of an app that exposes its actions to AI assistants through the Model Context Protocol. Connect your assistant to it, approve what it may do, and the assistant can call those actions when you ask.
Which AI assistants work with HackathonHost? Any assistant that supports MCP connectors, including Claude and ChatGPT. Assistants that only accept a pasted key can use an API key you create under Integrations.
Can the AI agent publish my event or email participants on its own? No. On HackathonHost an assistant cannot publish an event at all. It can send an email only after previewing it, showing you who it will reach and getting your explicit confirmation, and it never sends tickets.
Is it safe to let an AI agent read registration answers? Only with care. Participant-written text can contain instructions aimed at the agent. Use a read-only connection for that, and never one that can also make changes.
Do I need to know how to code? No. You connect the assistant once, approve its permissions in your browser, and then ask in plain language.



