# Knowledgebase: instructions for AI agents

This page belongs to the Knowledgebase web app, which only renders for a signed-in browser session.
If you are an AI agent, do not scrape it or try to sign in. Read it through the Knowledgebase MCP server,
which returns the same content as agent-ready text and respects the same permissions.

## Connect to the MCP server

- Endpoint (Streamable HTTP): https://knowledge.chaselubitz.com/mcp
- Auth: OAuth 2.1. An unauthenticated request returns 401 with a `WWW-Authenticate` challenge that starts the flow;
  a human approves access once in the browser. Clients that cannot do OAuth may send `Authorization: Bearer kb_pat_…`.
- Claude Code: `claude mcp add --transport http knowledgebase https://knowledge.chaselubitz.com/mcp`
- Claude / other MCP clients: add the endpoint as a custom connector and leave the OAuth client ID empty.

If you cannot add an MCP server yourself, ask the user to connect it, then continue with the steps below.

## Open this page through MCP

This URL does not identify a specific document. Ask the user for a workspace, or start with `search` / `list_entities`.

## General workflow

Find an entity with `search`, `query`, or `list_entities`; call `list_children` on it; then read items with
`read_resource` (default `text` format). Use `traverse` to follow relationships. Avoid `list_files` for
knowledge navigation, and use `read_file` → `edit_file` (with its `expected_version`) only for exact Markdown edits.