Docs · How to

How to: connect an AI client

A full walkthrough of the Connect panel — generating scoped tokens, copying the client config, verifying, and managing tokens over time.

The Connect panel is where NIBWP and your AI client shake hands. This is the first thing you will use and the thing you will return to whenever you add a new agent or a new team member. This guide walks through it end to end: generating a token with the right scope, copying the connection config for your client, and verifying the link — plus how to manage, rotate and revoke tokens as your setup grows.

TL;DR

In the Connect panel you generate a scoped application-password token and copy a ready-made connection snippet for Claude Code, Cursor or Claude Desktop. Each client gets its own token so you can revoke one without the others. Verify by asking the agent to list recent posts.

Open the Connect panel

In your WordPress admin, find NIBWP › Connect. This panel has three jobs: create tokens, show connection snippets for each client, and let you manage existing tokens. Think of it as the front door — every agent comes through here.

Step 1 — Generate a token

A token is an application password with a scope attached. Click to create one, give it a recognizable name (“Claude Code – staging”), and choose its scope. For a first connection, choose read:

ScopeGrantsUse for
readSearch and fetch contentFirst connections, research, audits
writeCreate and update draftsContent and data workflows
publishChange post status to liveTrusted, reviewed automations
adminSettings and usersAvoid unless truly necessary
Name and scope every tokenA token called “untitled” with admin scope is a future incident. Name each token for its client and purpose, and give it the smallest scope that works. You will thank yourself at audit time.

Step 2 — Copy the connection config

The panel generates a ready snippet for each client. Pick yours:

terminalbash
claude mcp add --transport http nibwp \
  https://yoursite.com/wp-json/nibwp/v1/mcp \
  --header "Authorization: Bearer YOUR_TOKEN"
~/.cursor/mcp.jsonjson
{
  "mcpServers": {
    "nibwp": {
      "url": "https://yoursite.com/wp-json/nibwp/v1/mcp",
      "headers": { "Authorization": "Bearer YOUR_TOKEN" }
    }
  }
}
claude_desktop_config.jsonjson
{
  "mcpServers": {
    "nibwp": { "command": "npx", "args": ["mcp-remote", "https://yoursite.com/wp-json/nibwp/v1/mcp"] }
  }
}
Keep tokens out of committed filesReference the token from an environment variable, never paste it into a file you commit. A leaked token is a door left open — though you can always revoke it from this same panel.

Step 3 — Verify the connection

Restart or reconnect your client, then ask the agent to do something read-only. If it returns structured results, you are connected:

claude > nibwp
You: List my 5 most recent posts.\nnibwp/wp-list-posts { limit: 5 }\n5 posts returned - connection verified

Managing tokens over time

As your setup grows, the Connect panel is also where you maintain it:

  • One token per client — so you can revoke or rescope one without disturbing the others.
  • Rotate on change — reissue a token when a team member leaves or a client offboards.
  • Revoke instantly — if a token is exposed, delete it here and the agent loses access immediately.
  • Scope up deliberately — start read, move to write when ready, and reserve publish and admin for trusted automations.

FAQ

Which transport should I use?
HTTP is the recommended transport. The legacy SSE transport still works but is deprecated for new connections.
My cloud client cannot reach the site.
Cloud clients (Claude, Cursor connectors) reach your site from the internet, so the endpoint must be public over HTTPS. Claude Code runs locally and can hit localhost or staging.
Can I connect more than one client?
Yes — generate a separate token per client and connect each. Independent tokens mean independent control.

Connect your first client

Generate a scoped token and link Claude or Cursor in minutes.

Get NIBWP

Connect your first agent.

Follow the quickstart, or book a demo and we'll walk you through it.