Content hub: news feeds, contacts, moderation, photo tags and AI assistants

Add feeds under Publishing > News feeds: a news site, a blog, a YouTube channel, a Mastodon or Bluesky profile. Paste the feed address, or the address of a page that offers one - the feed it advertises is found and kept. Only addresses that start with https:// and are on the public internet are read, and every read (and every redirect) is checked again.

Each feed is read about every hour (Settings > Content hub, 15 minutes to a day). A feed that keeps failing is read less often and switched off after 20 failures in a row; the reason shows on the list. A feed larger than 2 MB, or slower than 10 seconds, is not read.

New items wait under Publishing > News feeds > Feed items. Nothing shows until someone with Edit on the News area presses Show; Do not show keeps an item out for good. Titles and summaries are cleaned when they arrive: no pictures, links, scripts or frames, only simple text formatting.

Approved items are on the public page /feeds ("From around the web", never offered to search engines) and on a card of the member home page. Settings > Content hub can keep the page to signed-in members. Items are removed after 180 days unless you choose otherwise.

Our own feeds

Feed readers can follow your news at /news/feed.xml and your upcoming public events at /feeds/events.xml. Members-only, group and invitation events never appear in the events feed.

When switched on, every new or changed forum topic or reply, post, comment, class note and business listing is checked within a minute:

• With Use our AI service on (Settings > Content hub) and the AI assistant set up (Settings > AI assistant), your AI provider judges the text. Only the text is sent, never who wrote it; email addresses and phone numbers are replaced first. It uses the same monthly allowance as the writing assistant.
• Otherwise, or when the AI service does not answer, your own rules decide: a word list (whole words), more links than you allow, or text written almost entirely in capitals.

A flagged item goes to Community > Reports with the reason, where a moderator decides as for any report. A check never deletes or hides anything. Every check is listed under Community > Reports > Moderation log, with the reason and whether the AI service or the rules decided.

The MCP server lets an AI assistant - Claude Desktop, Claude Code, or any other client of the Model Context Protocol - look things up in your site. It can only read. Switch on both MCP server for AI assistants and Developer API under Settings > Modules.

What the assistant can do, each only when its token has the scope and the token's owner may still see it in the admin:

Connecting an assistant

1. Make an API token under System > API tokens with only the scopes the assistant needs, and a short expiry. Copy it (it starts adk_).
2. The server address is https://your-site/mcp. Every request sends the token as Authorization: Bearer adk_....

Claude Code:

claude mcp add --transport http alumdeck https://your-site/mcp --header "Authorization: Bearer adk_..."

Claude Desktop (Settings > Developer > Edit config, then restart). It reaches the server through the small mcp-remote bridge, which needs Node.js on your own computer - nothing on the server:

{
  "mcpServers": {
    "alumdeck": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://your-site/mcp", "--header", "Authorization:${ALUMDECK_TOKEN}"],
      "env": { "ALUMDECK_TOKEN": "Bearer adk_..." }
    }
  }
}

Other clients that speak MCP over HTTP (streamable HTTP) with custom headers take the same address and header, for example VS Code:

{ "servers": { "alumdeck": { "type": "http", "url": "https://your-site/mcp", "headers": { "Authorization": "Bearer adk_..." } } } }

Each token may make 60 requests a minute (Settings > Content hub). Every tool call and every read is listed under System > AI assistant log for full administrators: which token, what was asked and how many records came back - never the answers. Revoke a token under System > API tokens and the assistant stops at once.

Remember that what the assistant reads goes to its provider. Give tokens only the scopes needed, and mention AI assistants in your privacy notice if you use them.