AI tools and MCP
Connect your AI tools to Wyndlass
Wyndlass has an MCP (Model Context Protocol) server, so Claude and other AI tools can do what the CMS does: find screens, build content, schedule campaigns and report on what played. They sign in as you and get your permissions, never more.
The Wyndlass MCP server
Server address
https://mcp.wyndlass.ai/mcp- Connection
- Remote, over Streamable HTTP
- Sign-in
- OAuth, as yourself, with your Wyndlass account
- Tools
- About 150, in eight toolsets you choose from
- Works with
- Claude, ChatGPT, Cursor and VS Code, and other AI tools that can add a remote MCP server with OAuth sign-in
What it can do
About 150 tools, in eight toolsets
The tools cover what the CMS does. You pick the toolsets when you connect, and the tool sees only the parts of each that your roles allow.
Fleet
Find screens by place, type or status, see what one plays and why, take a screenshot, restart the player, pair new screens, and handle alerts and maintenance.
Content
Search the library by what pictures show, review held media, build playlists, draft and check layouts, themes and fonts, and import apps you build, once Imported apps is on.
Scheduling
Draft, publish and pause campaigns, aim them at saved audiences, and forecast where and how often they play.
Data
Create data tables and update their rows, such as prices or today’s specials, so the content that shows them changes.
Analytics
Ask Lookout what played: plays by day, item, campaign or screen, as screens reported them, with how complete the data is.
Advertising
With the media owner add-on: advertisers, ad campaigns, availability, booking against forecasts, and delivery reports.
Agents and tasks
Set up tasks that run on a schedule or an event, such as a nightly restart, start them, read their runs, see what an agent may do at a place, and write your organization’s skills.
Admin
Organize places, invite members, read the audit log, and check approval policies, API clients and AI credits.
Always included: who you are acting as, what is waiting for approval, undo for recent changes, and your organization’s skills.
Examples
Ask the way you would ask a colleague
A few typical requests, what the tool does with each, and the rule that keeps it in check.
Fleet
“Checkout 2 at Fargo South looks frozen. Take a screenshot, then restart the player.”
The tool finds the screen, gets a screenshot of what it shows, and sends the restart.
Commands keep trying for 24 hours and never run twice, and each is in the audit log as yours.
Content and Scheduling
“Make a playlist of the new spring photos and play it on the lobby screens from Monday until the end of April.”
It finds the photos by what they show, builds the playlist and drafts a campaign with those dates. Before publishing, it says how many screens the campaign reaches.
Publishing is confirmed first, and where an approval policy holds it, nothing goes live until an approver says yes.
Data
“The oat latte is $5.25 from today. Update the menu boards.”
It changes the price in the menu’s data table, and the menu boards that show that table pick it up.
Every save is a version, so the old price is one undo away.
Analytics
“How many times did the spring campaign play last week, by store?”
It asks Lookout for proof of play, as screens reported it, and says how complete the figures are if some screens have not reported yet.
It sees analytics only for the screens you may read them for.
Agents and tasks
“Restart the players on the drive-thru screens every night at 3 a.m.”
It sets up a task on a schedule in the site’s time zone, aimed at the screens tagged drive-thru. Every run is recorded, with each change it made.
The task’s runs act as you, with your permissions.
Admin
“Who changed the lobby layout yesterday? Put it back the way it was.”
It reads the audit log to find the change and who made it, then previews the revert, including anything changed since.
A revert always shows its preview first, and applies only once you confirm.
Illustrative examples. Names and numbers are made up.
How to connect
Add the server, choose what it may use, sign in
You need a Wyndlass account: the AI tool works as you.
Add the server
In your AI tool, add a remote MCP server at https://mcp.wyndlass.ai/mcp. The steps for Claude are below.
Choose what it may use
Wyndlass opens in your browser, names the app that is asking and says whether that name is verified, then lists the eight toolsets. Untick any the tool should not have.
Sign in as yourself
Sign in to Wyndlass the way you usually do. If you belong to more than one organization, choose the one the tool should work in.
Ask in your own words
The tool gets the toolsets you allowed that your roles cover, at the places you can reach. You can disconnect it in your AI tool at any time.
Steps for your AI tool
Claude on the web
- In Claude, open Customize, then Connectors, and choose Add custom connector.
- Paste the server address and choose Add.
- Choose Connect, and sign in to Wyndlass.
- In a chat, turn Wyndlass on from the + menu, under Connectors.
On Team and Enterprise plans, an owner adds the connector first, in Organization settings under Connectors. Each member then connects with their own Wyndlass account.
Claude Code
Add the server from your terminal:
claude mcp add --transport http wyndlass https://mcp.wyndlass.ai/mcpThen run
/mcpin Claude Code, choose wyndlass, and follow the sign-in steps in your browser. Add--scope userto use it in every project.Your organization’s skills then show as slash commands, such as
/mcp__wyndlass__price-change.To build an app of your own, add the app SDK’s skill to a project with
npx @wyndlass/app-sdk skill install, then ask for the app you want. More on building your own apps.Wyndlass warns that access is going to an app on your computer: that app is Claude Code.
Other AI tools
Add a remote MCP server (your tool may call it HTTP or Streamable HTTP) with the server address, and sign in when the tool asks.
The tool needs to support OAuth sign-in: the MCP server signs in people, and does not take API keys. For your own software, use the GraphQL API with an API client instead.
About the API
MCP server
Works with your AI assistant
Wyndlass works with Claude, ChatGPT, Cursor and VS Code through MCP, so you can manage screens, content, campaigns and data from the AI assistant you already use. The Wyndlass docs have the steps for each.
Your assistant signs in as you and gets your permissions, never more. Approvals still apply, and every action is in the audit log as yours.
Works with
- Claude
- ChatGPT
- Cursor
- VS Code
And other AI tools that can add a remote MCP server and sign in with OAuth.
The control model
The same rules as everyone else
An AI tool connected over MCP goes through the same permission checks, approvals and audit log as a person in the CMS.
Your permissions, never more
The tool acts as you: your roles, at the places you can reach, within each place’s limits on where AI may act.
Only the toolsets you allowed
You choose them when you connect. Toolsets your roles don’t cover are left out.
Confirmed before it applies
Wide-reaching and irreversible changes, and every undo or revert, are previewed first and apply only when you confirm. If anything changed in between, nothing is written.
Approvals still apply
A change a policy holds is reported as waiting, never as done, until an approver decides. Nobody approves their own change.
Recorded as yours
Every attempt is in the audit log under your name, marked as made through MCP and by which AI tool, including those refused or held.
Some things only a person can do
Granting roles, changing where AI may act or approval policies, and anything that makes a secret, such as API credentials or a screen PIN. Asked, the tool says so and points you to the page in the CMS.
You
Publish the spring campaign.
Your AI tool
Before I publish it: the spring campaign reaches 24 screens at 6 sites, and Midwest’s approval policy holds it for an approver. Shall I go ahead?
- Preview
- Needs approval
You
Yes.
Wyndlass
Waiting for approval by 1 person. Nothing has been applied.
Your AI tool
It’s waiting for an approver at Midwest, who gets an email. It goes live on its screens once they approve it.
AI tools and MCP: common questions
What is MCP?
MCP, the Model Context Protocol, is an open standard for connecting AI tools to other software. The Wyndlass MCP server gives AI tools about 150 tools that cover what the CMS does.
Which AI tools can connect?
Claude, ChatGPT, Cursor and VS Code, and other AI tools that can add a remote MCP server and sign in with OAuth. This page has the steps for Claude on the web and Claude Code, and the Wyndlass docs have them for each tool.
Do I need a Wyndlass account?
Yes. The AI tool signs in as you and works with your permissions. Access to Wyndlass is by invitation during launch.
Which plans include the MCP server?
Growth and Enterprise. The pricing page shows what each plan includes.
Can the AI tool do more than I can?
No. It gets your permissions at your places, only the toolsets you allowed, within each place’s limits on where AI may act. Some changes, such as granting roles, only a person can make, in the CMS.
Can it follow our own rules and routines?
Yes. It can read your organization’s skills, the guides your owners and admins write for its AI, such as how a price change is checked, and Claude Code offers each one as a slash command. A skill never widens what the tool may do.
Can it build apps for our screens?
Yes, in Claude Code with the Wyndlass app SDK’s skill: it builds the app with you on your computer, then imports, places and schedules it through the MCP server, asking before each. A person approves each version in the CMS before any screen plays it.
What does Wyndlass receive from my AI tool?
The requests it makes to the tools, with their inputs. The rest of your conversation stays with the AI tool. What the tools send back goes to the AI tool, and to its provider under your agreement with them.
How do I disconnect it?
Remove Wyndlass from your AI tool’s connectors at any time. If your access to Wyndlass ends, the tool’s access ends with it.
How is this different from the assistant in the CMS?
The assistant is built into every page of the CMS. MCP gives the AI tools you already use the same actions, under the same rules.
Can our own software use the MCP server?
Software should use the GraphQL API, with an API client: it gets its own role at a place and credentials that expire. The MCP server is for AI tools acting for a person.
Try it with the AI tools you already use
Access is by invitation during launch. Request a demo, and tell us which AI tools your team uses.