Skip to content

Let an AI app work on your sites

ReDock Web serves one MCP door for the whole workspace:

https://app.redock.xyz/mcp

Point Claude, Claude Code, Cursor or any MCP client at it. The first connection lands on a consent page here, and nothing reaches a site before somebody approves it.

  1. Add the address above to your MCP client as a remote server. What that looks like depends on the client.

  2. The client sends you here. The consent page names the app, the workspace it wants, and who you are signed in as. If ReDock has not verified who is behind the name the app registered itself with, the page says so in as many words.

    A consent page headed Connect Claude Desktop. It says the app wants to work on the sites in Tidewater Studio through ReDock Web, then a warning strip saying the app registered itself under that name and ReDock has not verified who is behind it. Below, a Which workspace list with two workspaces and the first one selected.

  3. Pick the workspace. The app reaches that workspace’s sites and no others. Approve it again elsewhere to give it another workspace.

  4. Pick the level. Under it, What approving means says the four things that follow.

    The Allow it to section of the consent page: four levels, View which is read-only, Draft which adds creating draft posts and pages, Edit which covers content, plugins, users and options on every paired site, and Full. Under them a line saying that whatever level is picked, the workspace’s AI policy does not allow publishing or trashing content, then Approve and Deny buttons.

    • Access is for this app only, in this workspace only, at the level you pick, and a level never widens what a site’s own plugin allows.
    • Every change it makes is recorded under its name, next to yours, with your role at the time.
    • Revoking it takes effect on its very next call.
    • If you leave the workspace, the approval goes with you.
  5. Approve. The app appears under Connected apps with its level, its client id, the date and a Revoke.

    The Connected apps page: three apps, two named Claude Desktop and one Claude Code, each with the level it was approved at, its client id, when it connected and when it was last used, and a Revoke button. Underneath, the workspace’s MCP endpoint, https://app.redock.xyz/mcp.

A per-site address reaches that one site and nothing else in the workspace:

https://app.redock.xyz/mcp/s/<fingerprint>

It is on the site’s own AI tab. Approving an app at that address gives it that site, and the rest of the workspace does not exist for it.

The consent level says what an app may be trusted with. The Usage policy says what any AI app may do on its own, whatever level it holds.

The Usage policy page. A line at the top says an AI app may not do these on its own: publish or trash content. Below, seven ticked or unticked switches with a sentence each, and a Save AI policy button.

Switch What it covers
Publish or trash content Making a draft live, taking a live page down, or moving something to the trash. Writing and editing drafts is unaffected.
Change what the front page is Pointing the site at a different home page, or switching between a static page and the blog.
Switch the theme Activating a different theme, which changes the whole site’s face at once.
Install, remove, update or toggle plugins Adding a plugin, deleting one, updating one, or turning one on or off.
Change site settings Writing WordPress options: the site title, permalinks, the plugins’ own settings.
Run database writes SQL that changes rows, and a real search-and-replace across the tables. Counting what a replace would change is unaffected.
Create users or change roles Adding a WordPress account, or moving somebody to a different role.

Switch one off and an AI app is refused it at the door and told to ask a person. The app is told which switches are off when it connects, not when it fails.

It does not stop you. Your team still does all seven from these pages, and so does anything you queue on the Jobs page.

A single site can be made stricter than the workspace, on its own AI tab. The bottom of the Usage policy page lists the sites that tighten it.

A script cannot sit through a consent page, so it carries a token instead.

The API tokens page: three tokens with their prefixes, levels, which sites they reach, when they were made, when they were last used, how many calls they have made and when they expire, two marked Live and one Expired. Under them, a Create a token row with fields for a name, the level, which sites and an expiry.

  • Allow it to is the same four levels the consent page offers.
  • Which sites is every site in the workspace, or one of them.
  • Expiry is set when the token is made.

A token acts as you, with your role as it stands on each call, and never above the level you pick. If you leave the workspace or your role changes, the token loses exactly what you lose, on its very next call. A token meant to outlive a person is made by the owner.

Use it as a bearer token against the same door:

Authorization: Bearer rdw_...

against https://app.redock.xyz/mcp, or a site’s own address when the token is for one site. The four export files take one too: /activity.csv, /activity.json, /reports.csv and /reports.json.

Revoke is next to every app and every token. The next call is refused. Not the next hour, and not after a session expires.

Removing somebody from the workspace disconnects their AI apps in the same moment.

Every refusal is on Activity, with the code and what was asked.


Open Connected apps in ReDock Web.