Skip to content
Agent Abilities for MCP Agent Abilities

Getting started

Four steps take you from a fresh install to your first agent call. Everything starts off, so you decide what the agent can reach before it connects.

  1. Turn on what you want. Open the Agent Abilities for MCP screen and enable only the abilities you want the agent to have. Everything starts off.

  2. Copy your endpoint. On the Connection tab, copy your site’s MCP endpoint.

  3. Connect your AI client. The simplest path is OAuth: paste the endpoint into your client and approve once in the browser. Or point a dedicated low-privilege user at an Application Password.

  4. Check the connection. Use the connection check on the Connection tab to confirm the endpoint is reachable from your server.

Your first call, start to finish

Section titled “Your first call, start to finish”

If you want one concrete run to build confidence, start with a read-only ability so nothing on your site can change:

  1. Enable one low-risk ability. Turn on Site-wide search, which only reads and spans every post type. Leave everything else off.

  2. Connect one client. Point a single MCP client (Claude Desktop or Cursor, for example) at your endpoint and connect it as a scoped user.

  3. Ask it a real question. Have the agent search your own content, for instance “find my posts that mention pricing”. It runs the search ability you enabled and nothing else, because nothing else is switched on.

  4. Watch the call land in the log. Open the activity log and find the row for that call: the acting user, the ability name, and the argument keys, values never stored. Denials show up here too, so you can see exactly what the agent did and did not do.

Once that round trip feels right, enable the next ability. You add reach as you build trust, one ability at a time.

What happens on that first call

Section titled “What happens on that first call”

When your client calls an ability, the plugin checks it against the bound user’s capabilities before it runs, then writes the call to the audit log, the acting user and the argument keys included. If the user is not allowed to run that ability, the call is denied and the denial is logged too. You add reach as you build trust, one ability at a time, not all at once.

Next steps

Section titled “Next steps”