Users, agents and integrations

Clients that act as you: OAuth grants, capabilities and disconnecting.

Three different things can connect to Elixir MCP, and the difference is not technical trivia — it decides what "me" means, whose data comes back, and who pays for the calls.

The short version:

You

When you connect Claude — or any MCP client — to https://elixir.poapkings.com/mcp, you are connecting as yourself. You sign in with the email on your account, and from then on the tools know who you are.

That last part matters more than it sounds. Ask "how am I playing?" and your agent does not need to look you up first: your connection already knows your primary player, your alts, the friends you follow, and your clan. Omit the player tag and it means you.

You can only have one self, so there is nothing to configure. This is the connection almost everybody wants.

Signing in to other Elixir products with this account

Elixir Clan (and, soon, Elixir Drop) can sign you in with Elixir: the same consent page, plus one extra line, Know your email address, which only an app that asks for it is ever shown. That is how those products know you are the same person you are here. It is listed on Connections like any other client, and disconnecting it there ends its access; the address itself is never shown to an MCP client or an agent. See Protocol → Signing a person in.

Your first question

On Connections, Try asking… offers questions based on your recorded data. Copy one and paste it into the AI client you connected. A profile can support a snapshot summary; battle history can support a review, deck comparisons or two-week comparisons. A clan-war question appears when a clan you added has a recorded war week. You can view the full question before copying it.

These are curated starting points. They ask your client to check freshness, sample sizes and missing data. If capture has not arrived yet, the panel explains what to do next and refreshes while open. Copying a question does not call the MCP service or count as an answered question.

An agent

An agent acts for a clan, not for a person.

It is a separate principal that you own: its own identity, its own key, its own event feed, its own notion of who it is. What it does never lands in your history, and what you do never shows up as its.

The reason to want one is that a clan is not a person. If a Discord bot connected as you, then "who do you track?" would answer with your personal player list, and everything it did would be recorded as something you did. An agent has no personal players to leak, because it has no self — its subject is the clan.

You can create an agent for any clan you already record, at any tier. It is not a paid feature. It spends your daily call budget, and it carries its own tier — an admin's agent is not an admin.

Agents serve many people

A clan agent talks to a whole Discord, so "me" is a different person every time. MCP has no way to say which human is asking, so the agent tells us: it passes whatever id its own surface has — discord:1234, signal:…, anything — and the first time we don't recognise someone, it asks who they are in the clan and remembers.

That mapping only picks a default player. It grants nothing: recorded data is readable by every account either way.

An integration

An integration has no "me" at all. It reads the corpus on behalf of its own users, naming what it wants on every call.

Elixir Drop is the example: it looks up whoever is playing, in whatever clan they happen to be in, and has no relationship with any particular clan of its own.

Integrations use the REST Integration API, with an admin-issued server key, explicit permissions and their own capacity. They do not inherit the human sponsor's admin powers or personal quota.

Which do I want?

• Playing Clash Royale and curious about your own history → connect as yourself. Nothing else to do.
• Running a clan and want a Discord bot, or Claude, doing clan work → make an agent for the clan.
• Building an app for other people → an integration.

One connection at a time

You can hold more than one — a personal connection and an agent for your clan — but keep them in separate sessions, not both switched on in the same chat.

They publish the same tool names, so an assistant with both cannot tell which one you meant, and the two answer from different subjects. In Claude that means a dedicated project for clan work with only the agent connector enabled.

Each connection has its own URL, so a credential presented at the wrong one is refused rather than quietly answering about the wrong subject.

Where things are in the console

The rail groups the console into what you are reading, what we record for you, who can call on your behalf, and how the service is running.

Making one

Create clan agents under Connections → Agents. Platform integrations are managed under Admin → Integrations. Both show a newly issued key once; only its hash is stored. See the integration guide for REST resources, permissions and automatic recording enrollment.