All topics / About Everything / Getting started / Connect your agent
Check that your Everything connection works
Confirm your AI assistant actually read a public page, then try browsing and comparing related ideas.
Your AI assistant should be able to read an actual Everything page before you rely on its answers. A connection listed in settings is only the first step.
1. Find Everything in your agent
Open your agent's list of MCP servers, tools or connectors. Look for Everything. Depending on the client, you may need to enable it for this conversation.
Typical public tools are get_page, list_pages, search_pages and get_related_pages. Some clients also expose a public_status check.
2. Ask for one specific article
Copy this into your assistant:
Use the Everything MCP
get_pagetool to read/topics/software-and-agents. Give me the actual title and link to one collection mentioned on that page.
The answer should name Software & Agents and link to Beyond the Application. Inspect the tool call if your client shows it. A plausible answer with no tool call is not proof the connection is working.
3. Follow the argument
Then ask:
Read
/beyond-the-application. What are its four main essays? Recommend a reading order, and link each title.
Compare the response with the collection page.
4. Explore on your own terms
Try:
Search Everything for dynamically generated interfaces. Read the most relevant article, show the URL, and explain how it relates to the disaggregation of software.
Or ask your agent to browse related pages, explain unfamiliar terminology, or contrast the text with other sources. You can ask it to distinguish the author's claims from the assistant's own interpretation.
What successful access means
Visitors can read and search published content, but not edit or delete it. No Everything account is required. An agent might list tool names it cannot actually use; that does not change visitor permissions.
Search indexes may lag newly published pages. If you already know a URL, reading that page directly is the best check.
For errors, see Troubleshooting. For how a public tool request is handled, see Access and privacy.