AI and MCP API: Connect Claude, ChatGPT and Cursor

SolumeCore has an AI API built on the Model Context Protocol (MCP). It lets an AI assistant you already use, such as Claude, ChatGPT, Cursor, VS Code or n8n, manage your own game servers, apps and VPS in plain words: read the console, restart, edit a config line, install a plugin, make a backup. You decide which servers it can see, what it may do in each area, whether every change needs your approval, and when its access expires. Everything it does is logged.

This guide explains how to connect an assistant, how permissions and approvals work, which tools are exposed and which safety nets exist. It also covers the Copilot, the AI chat that can appear inside your server pages. SolumeCore also publishes a step-by-step guide for connecting your AI.

Important: the AI API is a beta. You must accept the beta terms before creating your first token, and during the beta only your own servers can be connected, not servers shared with you. The Copilot is a separate feature that is switched off by default: it appears only if SolumeCore has enabled it, so if you do not see it, it is not available on the platform yet.

The pieces

The rest of this guide is about the AI API unless it says otherwise.

Connect your AI

The MCP address is https://api.solumecore.com/mcp. There are two ways to authorize an assistant.

One click with OAuth (Claude on the web and app, ChatGPT): add a custom connector named SolumeCore with the MCP address, press Connect, and SolumeCore brings you to the panel to pick servers and permissions. No token to paste. In Claude: Settings → Connectors → Add custom connector. In ChatGPT: Settings → Apps & Connectors → Advanced, turn on Developer mode, create the connector with OAuth authentication.

With a token (Claude Desktop, Claude Code, Cursor, VS Code, n8n, anything else):

1. Open Account → AI & copilot, accept the beta terms if you have not, and click New AI token.
2. Walk through the five steps (below), click Create token and copy the token. It starts with slm_ai_ and is shown only once.
3. Open the Connect tab, pick your assistant and copy the configuration, replacing YOUR_TOKEN with the token.

claude mcp add --transport http solumecore https://api.solumecore.com/mcp \
  --header "Authorization: Bearer slm_ai_xxxxxxxxxxxxxxxx"

{
  "mcpServers": {
    "solumecore": {
      "url": "https://api.solumecore.com/mcp",
      "headers": { "Authorization": "Bearer slm_ai_xxxxxxxxxxxxxxxx" }
    }
  }
}

{
  "servers": {
    "solumecore": {
      "type": "http",
      "url": "https://api.solumecore.com/mcp",
      "headers": { "Authorization": "Bearer slm_ai_xxxxxxxxxxxxxxxx" }
    }
  }
}

For n8n, add an AI Agent node with an MCP Client Tool, set the endpoint to the MCP address, the transport to HTTP Streamable and the authentication to Header Auth with name Authorization and value Bearer YOUR_TOKEN. For automations, choose Approve in the panel so changes wait for you. Claude Desktop needs a small claude_desktop_config.json block (it uses Node.js); the Connect tab shows the exact text.

Any MCP client that speaks Streamable HTTP works, and the older SSE transport is available at /mcp/sse. A raw request looks like this:

curl https://api.solumecore.com/mcp \
  -H "Authorization: Bearer slm_ai_xxxxxxxxxxxxxxxx" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

An unauthorized call gets 401 with the discovery information MCP clients need for OAuth; a token used from a disallowed IP or an inactive account gets 403. Tool names and error messages are in English.

Create a token

The New AI token assistant has five steps.

1. Name: something you recognize in the list and the log, such as "Claude Desktop".
2. Servers: All my servers (including ones you buy later) or a selection of game, app and VPS services.
3. Permissions: start from a preset and fine-tune by area.
4. Safety: confirmation mode, backup before big changes, changes per day, expiry and allowed IPs.
5. Review and Create token.

Permissions

Each area has a level: No, Read, Changes or Full, and each level includes the one before it. Three presets set all areas at once:

The areas are Servers and metrics, Power, Console and logs, Files, Backups, Players, Plugins and mods, server.properties, Scheduled tasks and VPS. A Read token can look but never change; destructive tools need Full in their area.

Confirmation mode

A proposal that you do not answer expires after 15 minutes. Destructive actions always need confirmation, and for a token that can do them the panel recommends Approve in the panel, because in chat mode the AI itself confirms and a file or log with hidden instructions could push it to confirm something by itself. By default SolumeCore also requires panel approval for destructive actions even on tokens set to chat mode; this is a platform setting that SolumeCore can change.

Other safeguards

• Backup before big changes: before restoring a backup, installing or updating plugins or changing server.properties on a game server, a backup is taken first (on by default).
• Recoverable deletes: files an AI deletes move to an IA-deleted-files folder and can be restored for 3 days, from the Deleted files tab or by asking the AI; after that they are erased.
• Changes per day: a daily cap on changes (100 by default, up to 2000). Readings do not count. At the cap the AI can still look but not change.
• Expiry: 1, 7, 30, 90 or 365 days. There are no eternal tokens.
• Allowed IPs (optional): up to 20 addresses or ranges, one per line, for tokens used from n8n or your own scripts. Empty means from anywhere.
• Rate limits: 120 calls per minute per token, and 20 changes per minute.
• Up to 25 active tokens per account. Revoking one disconnects the AI immediately and cancels its pending proposals.

Tools the AI can use

The server exposes 45 tools. Names are English and snake_case, as MCP clients expect. Destructive tools are marked.

Things you can ask: "Why did my Minecraft server crash last night?", "Back up Survival and then restart it", "Set max-players to 50", "Search the log for plugin errors", "Install EssentialsX and LuckPerms", "Add Steve to the whitelist".

Approvals, activity and audit

• Approvals tab: pending proposals with the server, the action and, for files, the diff. Approve runs it; Reject cancels it. You also get a notification, "Your AI wants to make a change", when a proposal is waiting.
• Activity tab: every call with the time, token, tool, server and result (OK, Error, Proposed, Denied, Limit or Waiting), its parameters with secrets hidden, and the source IP. It is kept for 180 days, and a chart shows calls, changes and errors for the last 14 days.
• Deleted files tab: what AIs deleted, with Restore and Erase now.
• Want a message in Discord, Slack, Telegram, Teams, email or your webhook each time an AI changes something? Use the ia.accion event; see Integrations and notifications and Webhooks and account events.

The Copilot in your server pages

When SolumeCore enables it, a Copilot button opens an AI chat inside a server. It reads the console, logs and files of that server, explains what is happening in plain words and proposes changes that you review and apply with Apply or throw away with Discard. Nothing changes until you apply it, and a backup is made first. There is a daily question limit; the panel shows "X of Y today". Players' IP addresses are masked in what is sent to the AI by default. You can hide the Copilot from all your servers with the toggle "Show the copilot in my servers" on the AI & copilot page. Like any AI, it can be wrong.

Privacy and responsibility

The AI acts with the permissions you give each token, and you are responsible for what you approve. What the assistant reads to answer you (status, console and log lines, the contents of files it opens) is processed by the AI provider behind that assistant under that provider's own terms (for the Copilot, the provider SolumeCore uses). It does not see your password, payment details or the RCON password. To keep something out of reach, give the token fewer servers or the Look only preset.

Related documents

• Which SolumeCore API to use
• Integrations and notifications
• Webhooks and account events
• Game server panel
• Backups and retention
• Account security

FAQ

What is the SolumeCore MCP server?

It is an endpoint, https://api.solumecore.com/mcp, that lets MCP-compatible AI assistants list, read and manage your own servers using tokens or OAuth, with permissions, approvals and an audit log that you control.

Can the AI do something without my approval?

By default no: each change is a proposal you approve in the chat or the panel, and destructive actions always ask. If you choose "Don't ask for normal changes" on a token, routine changes such as restarts and file edits run immediately, and every call is still logged.

What happens if the AI deletes something important?

Deleted files go to the IA-deleted-files folder of the server and can be restored for 3 days. You can also keep Backup before big changes on so a backup is taken before risky actions.

Does it work with servers other people shared with me?

Not during the beta. Only servers on your own account can be connected.

How do I cut an assistant's access?

Revoke its token in Account → AI & copilot. It stops working at once and pending proposals are cancelled. Tokens also expire on their own and can be limited to certain IPs.

Why can't I see the Copilot?

It is switched off by default and only shows if SolumeCore has enabled it. You can also have hidden it with the toggle on the AI page, and it may be unavailable for your plan or paused for the day.