Troubleshooting your Everything connection

If your AI client has trouble connecting to Everything, start with the public reading checks below. No Everything account or password is required.

I entered the URL in a browser and saw an error

The MCP endpoint https://everything.emlynoregan.com/mcp/ is a protocol endpoint, not a web page. A browser GET is not a valid test of every MCP operation; the client normally negotiates protocol messages over HTTP. Use First tests or visit the human-facing Everything homepage.

Use only the address shown in the connection guides.

The MCP server won't appear

1. Confirm you used https://everything.emlynoregan.com/mcp/, including the /mcp/ path.
2. Ensure the client supports remote Streamable HTTP MCP servers rather than only local stdio.
3. Inspect the client's server list, permissions and connection-error message. Refresh/restart the client if necessary.
4. Verify the configuration format: Cursor uses mcpServers, VS Code uses servers, Codex uses mcp_servers in TOML.
5. In ChatGPT, custom connection availability depends on plan/workspace permissions; in Claude, the connector may need to be enabled for the chat.

The server appears, but my agent ignores it

Ask explicitly for Everything MCP's get_page, name a known path and ask the model to report the actual returned title. Some agents prefer their own knowledge or web search without a specific tool request. Verify tool invocation rather than accepting a plausible summary.

The client asks me to sign in

Published pages can be read anonymously. Select No sign in or No authentication if available. If your client requires a login and cannot offer anonymous access, that client may not support this public connection mode. Do not enter an unrelated account password.

Can I edit articles?

No. Visitor access is read-only, even when a client shows names of tools that are not available to the public.

Requests are slow or time out

Initial requests may involve cold application or storage dependencies. Avoid rapid repeated searches. Retry a read deliberately, or read a known page by route instead. Client/network/protocol overhead is distinct from storage time.

A page exists but does not appear in search

Direct get_page by known route is the right verification. Indexes and related-result metadata can lag publication. Incomplete or partial search coverage must not be interpreted as proof of absence. Try the known page URL directly instead of relying on search results.

Where to go next

• Connection guides — choose your AI client and set up the server
• Check the connection — verify a genuine page read
• Access and privacy — understand the public read-only connection

Browse Everything