Connect an AI coding tool to your WordPress site
The Unyson+ AI Assistant extension includes an MCP server, the open standard AI tools use to call other software. Connect your AI tool to it and you can say "build a draft About page with our story and team" in Claude Code, Cursor, VS Code, Windsurf or Claude Desktop, and it builds the page in your site's page builder: checked against the real element options, saved as a draft, and undoable.
This guide is the setup for each tool. The technical details (methods, errors, limits) are in the MCP server reference.
:::note Just want to chat inside WordPress? You do not need any of this for the ✦ AI Assistant chat panel inside WordPress; it finds its AI a different way. See Set up AI for Unyson+. :::
:::tip Does your tool offer a browser sign-in for MCP servers?
Then skip the password: add the server URL
https://example.com/wp-json/unysonplus-ai/v1/mcp as a remote MCP server, and the tool opens a page on
your site asking "Allow … to use this site?". Choose Read and write or Read only and press
Allow. See Web apps below. The password steps that follow work
with every tool.
:::
1. Create a connection password on your site
- In WordPress, activate AI Assistant (Beta) under Unyson+ → Extensions.
- Open Unyson+ → AI Assistant → Advanced → Outside AI programs.
- Name the connection after the tool (for example "Claude Code on my laptop") and press Create a connection password.
- Keep the page open: it shows the details once. Press Copy everything to copy the Server URL,
username, password and Authorization header (it starts with
Basic) in one go — paste it to your AI tool or straight to the AI that asked. Separate fields and a ready-made MCP config are under Other formats.
Access is switched to Read and write. Choose Read only on the same screen if the tool should only
look. Your site must use HTTPS (a local development site on localhost, .local or .test is
fine too).
In the examples below, replace https://example.com with your site and Basic YOUR_TOKEN with the
Authorization header you copied.
2. Add it to your AI tool
Claude Code
One command in a terminal (add --scope user to use it in every project):
claude mcp add --transport http unysonplus https://example.com/wp-json/unysonplus-ai/v1/mcp --header "Authorization: Basic YOUR_TOKEN" --scope user
Check it with claude mcp list: the server should show Connected. Then start claude and ask for
something.
Cursor
Add the server to ~/.cursor/mcp.json (all projects) or .cursor/mcp.json (one project):
{
"mcpServers": {
"unysonplus": {
"url": "https://example.com/wp-json/unysonplus-ai/v1/mcp",
"headers": { "Authorization": "Basic YOUR_TOKEN" }
}
}
}
Open Cursor Settings → MCP and make sure unysonplus is enabled, then use Agent mode in chat.
VS Code (GitHub Copilot agent mode)
Add a .vscode/mcp.json to your workspace:
{
"servers": {
"unysonplus": {
"type": "http",
"url": "https://example.com/wp-json/unysonplus-ai/v1/mcp",
"headers": { "Authorization": "Basic YOUR_TOKEN" }
}
}
}
Start the server from the file (VS Code shows a Start link above it), then open Copilot Chat in Agent mode; the site's tools appear under the tools icon.
Windsurf
Add the server to ~/.codeium/windsurf/mcp_config.json:
{
"mcpServers": {
"unysonplus": {
"serverUrl": "https://example.com/wp-json/unysonplus-ai/v1/mcp",
"headers": { "Authorization": "Basic YOUR_TOKEN" }
}
}
}
Refresh the MCP servers in Cascade's settings, then ask Cascade.
Claude Desktop
Claude Desktop connects to remote servers through a small local bridge that adds the header. It needs Node.js installed. Open Settings → Developer → Edit Config and add:
{
"mcpServers": {
"unysonplus": {
"command": "npx",
"args": ["-y", "mcp-remote", "https://example.com/wp-json/unysonplus-ai/v1/mcp", "--header", "Authorization:${UPW_AUTH}"],
"env": { "UPW_AUTH": "Basic YOUR_TOKEN" }
}
}
}
(The header goes through an environment variable because a space inside the arguments breaks on some systems.) Restart Claude Desktop; the tools appear under the tools icon in a new chat.
Any other MCP tool
Add a remote HTTP (Streamable HTTP) MCP server with the Server URL and an Authorization header set
to the value you copied. Most tools accept the config block the connection screen shows as-is.
Web apps: ChatGPT and claude.ai
ChatGPT connectors and claude.ai custom connectors connect through a browser sign-in (OAuth) instead of a password, which the site supports from AI Assistant 1.0.19:
- In the app's connector settings, add a custom connector (remote MCP server) with the URL
https://example.com/wp-json/unysonplus-ai/v1/mcp. Leave any client ID / secret fields empty: the app registers itself. - The app opens your site. Log in to WordPress if asked, choose Read and write or Read only, and press Allow.
- Back in the app, the site's tools are available in your chats.
These apps connect from the provider's servers, so your site must be public and on HTTPS: a site on your own computer cannot be reached. The app appears under Apps signed in with your account on the Outside AI programs screen, where Sign out disconnects it.
3. Try it
Good first requests:
- "What is on this site? Use site_info." (read-only, safe to try first)
- "Create a draft page called Pricing with three plans and a FAQ, then check it for problems."
- "Give the site a warm colour palette and friendly fonts." (Theme Settings changes are live; ask it to undo them if you do not like them)
New pages are drafts. Every change saves a revision first, and the tool can undo its own changes
(undo, undo_theme_settings, undo_change).
If it does not connect
Test the connection with the two curl commands in the
MCP server reference, then check its
troubleshooting table. The two most common
causes: the extension is not active (a 404), or the host strips the Authorization header (a 401 with
the right password), which one .htaccess line fixes.
When the tool has connected, Last used in the connections list shows the time. To disconnect a tool, press Revoke next to its connection (or Sign out next to an app that used the browser sign-in).