A Knowledgebase is written for agents as much as for people. Connect Claude, a coding agent or your own script, and it can search the pages, read them with how far each can be trusted, and suggest edits for a person to accept.
There are three ways in. Each is listed in the Connect an Agent dialog, with the address and the commands to copy.
Open Connect an Agent
Press the AI Knowledge tab in the footer.
Press Connect an Agent at the top of the view.
Anyone who can open the Knowledgebase can read the dialog. What a connection can then do depends on who it acts for (see What Agents Can Do).
Through the MCP Server
This is the way for Claude and any other AI tool that speaks the Model Context Protocol.
Copy the address under MCP Server: https://mcp.liveknowledge.app/mcp.
Add it as a connector or a remote MCP server in your AI tool: in Claude, open Settings, then Connectors, and add a custom connector with this address as its URL. Other tools take it wherever they ask for a remote MCP server or a server URL.
Your AI tool opens a liveknowledge page asking you to approve the connection. Approve it. The tool makes its own token, so there is nothing to paste.
Ask it something about your pages. It calls get_context first, then search_pages,
read_page and suggest_edit. Every tool is listed in The liveknowledge MCP Tools.
Through the Command Line
The liveknowledge command reads and suggests from a terminal or a script.
liveknowledge auth login
liveknowledge context Handbook
liveknowledge search Handbook when do we deploy
auth login signs you in once in the browser. Every other command takes the Knowledgebase's name (or its id, or its address). See The liveknowledge CLI.
With an API Token
Press Open API Tokens in the dialog. It opens the Tokens page in Settings.
Create a token and copy it. It is shown once.
Send it to the API as Authorization: Bearer <token>. See API Tokens and The Knowledge API.
A read-only token only reads, whatever is allowed in the Knowledgebase.
See It Connected
Once an agent has read or written this Knowledgebase, it has a card under Connected Agents in AI Knowledge.
- The name is the one the AI tool gives itself, or "CLI" for the command line.
- How it connects: "Through the MCP server", "Through the command line" or "With an API token", and when it was last seen.
- What it may do, as chips: Reads, Suggests Edits and, once an Editor allows it, Writes.
- Pages read this week, and the three it read most. Press one to open it.
A card stays for 30 days after the agent last read or wrote here.
Good to Know
- An agent acts for the person or token that connected it, and can open only what they can. A share role applies to an agent exactly as it does to a person.
- A new connection reads and suggests. Writing is allowed one connection at a time, by an Editor. See What Agents Can Do.
- A Knowledgebase kept Offline is in your browser only, so no agent can reach it.
Who Can Do What
| Action | Editor | Participant | Viewer |
|---|---|---|---|
| Open Connect an Agent, see Connected Agents | Yes | Yes | Yes |
| Connect an agent that reads | Yes | Yes | Yes |
| Connect an agent that suggests edits | Yes | Yes | No |
| Allow an agent to write, or stop it | Yes | No | No |
Was this article helpful?
Still need help?
Try Troubleshooting, or tell us what you were trying to do.