Integrations

Agents and AI tools

These docs as a machine-readable source, and what an agent can actually do with the platform today.

Every page on this site has a Markdown twin, because a good share of the traffic here is not human.

These docs, read by a machine

Four ways in, all generated at build time, none of them a separate product.

/llms.txt is an index of every page with its description, small enough to paste into a prompt. /llms-full.txt is the entire site as one Markdown file, and it is the one you hand to an agent.

The other two are for reaching a single page. Every page's source sits at /llms.mdx/docs/<path>/content.md, and every page header carries a Copy Markdown button that fetches it. Ask for any docs URL with an Accept: text/markdown header and you get the Markdown back instead of a page of HTML to scrape.

The old documentation served Markdown for every page and partners' agents came to depend on it. That is why it still works here.

Uses today

Configuration questions get answered without anyone reading the site. Point Claude, Cursor or Codex at /llms-full.txt and ask the question you would otherwise ask us. How do I set up a daily quest with a streak cap, what does hidden if locked actually do, which templates need an account linked first. The answers are all in there.

Reports arrive without your team building them. Hand an agent the Data API reference and a key, then ask for the thing you actually want. The fans who completed a quest last month but have not opened the app since, as a CSV. It writes the calls, pages the results and gives you the file. It needs nothing from us.

Both are read-only work on your side, and neither one needs a feature to be built first.

Building the fan experience with an agent

If you are putting FanBase into your own product rather than running the hosted app, the agent path is how the SDK is meant to be installed. The packages carry the files an agent reads, a bootstrap prompt and a thin agent skill, so the reference material arrives in the repo alongside the code.

Paste two prompts, and the order matters: sign-in first, questing second. Combining them produces something that looks finished and has a stubbed login. Integrate with an agent has both prompts and the list of things worth checking by hand afterwards.

An assistant in the admin panel

Exploring

The idea is that you would ask for the thing instead of filling in the form. Create a daily quest worth fifty XP with a streak cap, and it would draft it for you to approve. Show me quest completions by tag for last month, and it would pull the report.

It is under discussion rather than scheduled. If it would change how your team works, tell your FanBase contact: knowing a partner wants it is what gets it a date.

Agents connect by reading files, not by MCP

Not for this platform, today. An agent that reports otherwise is guessing, and there is nothing to install or connect.

The supported pattern is to give the agent files to read and an API key when it needs one. It works with any tool that can read a file.

On this page