Connect Tagboard MCP to your AI Assistant

Tagboard MCP lets you connect your AI Assistant directly to Tagboard Producer. Think of it as a remote control you operate by typing or speaking to your assistant. 

With it, you can ask things like "What's in the playlist for tonight's show?" or "Go live with my pregame show at 6:00PM. Loop the playlist for 30 minutes until kickoff." 

This also puts your AI assistant to work inside Tagboard Producer right alongside everything else your assistant can reach: your other tools, its built-in knowledge, and the data you connect to it. Combine them and one request could check a stat, read your rundown, and preview your graphic.

This article will cover:

1. Connecting the MCP to your AI Assistant
2. Running your first check
3. What you can ask it to do
4. Staying safe while live
5. Troubleshooting

Getting Started

You will need:

• A Tagboard login
• Your AI Assistant, signed in to the workspace where you want the connection
• A plan or workspace setting that allows custom connections

You do not need an API key, a tunnel, an administrator account, or anything running on your computer to set this up.

Choose your setup:

• Claude
• ChatGPT 
• Codex 
• Other MCP clients and manual setup

The connection address is the same for everyone:

https://mono.tagboard.com/mcp

Type the whole address, including /mcp at the end. If you paste it into a web browser you'll see an error. That's normal and doesn't mean anything is broken.

Claude

Claude Desktop or in a browser

1. Go to Customize > Connectors > Add custom connector.
2. Enter the following production name and URL:

   Suggested connection name	Tagboard MCP (Production)
   Server URL	https://mono.tagboard.com/mcp

3. If the connector wizard shows you authentication options, select Sign in Now and Use Claude's Published Identity (CIMD). Leave the rest of the Advanced Settings as is. Do not enter any client IDs or credentials.
4. Click Connect. A browser window will open up the Tagboard login page. Sign in with your Tagboard account and review the permissions. (In the Desktop app, you may see a prompt that asks, Finish Connecting a Connector? before opening Tagboard login. Continue only if you started the connection yourself.)
5. Back in Claude, turn on the connector for your conversation and you're ready to run your first check.

Note that organization owners may need to add it under Organization Settings > Connectors first. 

ChatGPT

ChatGPT in a Browser

1. Open your ChatGPT workspace
2. To enable a custom connection, first open Settings > Security and login > Developer mode, and enable it if your workspace allows it.
3. Open Plugins, choose Add or the + button and select the custom MCP connection option. 
4. Enter the following production name and URL:

   Suggested connection name	Tagboard MCP (Production)
   Server URL	https://mono.tagboard.com/mcp

5. Create the connection, then choose Connect if prompted. You'll be guided to log in to your Tagboard account, and review consent in the browser. 
6. Install and enable the resulting plugin, if prompted. Then start a new conversation, type '@' and select your Tagboard connection. Now you're ready to run your first check.

ChatGPT Desktop App

There are two distinct connection surfaces in some desktop releases:

• Hosted plugin/app: follow the browser setup above using the same account and workspace, then open the desktop app's Plugin/Apps and enable that connection in your conversation. If it does not appear, check whether that desktop release supports the hosted plugin.
• Local MVP Server: if settings exposes MCP servers or Plugins > MCPs > Add, follow the Codex desktop setup below. This configures the local host, not a browser-hosted connection. 

If the desktop app has no custom-connection control, use the browser path above. Do not enable Browser or Computer use developer settings to compensate; those controls serve a different purpose.

Codex

Codex Desktop App

1. Open Settings > Plugins > MCPs > Add and choose a custom MCP server.
2. Enter the production name and URL below, and select Streamable HTTP. Leave bearer token environment variable and both header sections empty. Click to save.
   

   Suggested connection name	Tagboard MCP (Production)
   Server URL	https://mono.tagboard.com/mcp

3. Keep OAuth registration Automatic, or select CIMD if offered. Do not force dynamic client registration (DCR).
4. Select Authenticate and sign in to your Tagboard account in the browser. If the app request a server restart, use its restart control, and start a new task. Now you're ready to confirm Tagboard tools are available, by running your first check.

Codex CLI and manual configuration 

For an installed Codex CLI< add the server once: 

codex mcp add tagboardProduction --url https://mono.tagboard.com/mcp

If adding it did not finish authentication, run:

codex mcp login tagboardProduction

Alternatively, add this to your existing ~/.codex/config.toml without replacing other entries, then run the login command:

[mcp_servers.tagboardProduction]
url = "https://mono.tagboard.com/mcp"

Use one setup method, not duplicate entries for the same server. codex mcp list shows configuration; /mcp in a new interactive Codex session shows available servers. Neither replaces a successful authenticated tool call.

Local desktop, CLI, and IDE clients can share the configuration on the same Codex host. This does not copy credentials into a separate cloud environment. See the official MCP guide.

Codex In a Browser

Do not paste the local TOML configuration into a browser chat or assume your computer’s connection is available to a cloud task. For ChatGPT Work in a browser, use the hosted ChatGPT plugin path above. A separate browser-hosted Codex coding environment needs its own supported integration controls; that path has not been validated for Tagboard. Use ChatGPT web or local Codex until it is validated.

Other MCP Clients and Manual Setup

Choose a remote server with Streamable HTTP and enter the production URL. The client must support OAuth discovery, authorization code with PKCE, and a client registration compatible with Tagboard’s Auth0 configuration. Let it discover the authorization server rather than manually constructing a login URL.

For an advanced client asking for these values:

Field	Production value
Resource / audience	https://mono.tagboard.com/mcp
Protected-resource metadata	https://mono.tagboard.com/.well-known/oauth-protected-resource/mcp
Authorization-server issuer	https://tagboard.auth0.com/
Basic read scopes	openid producer:read
Refresh access, if supported by the client	offline_access

Scopes are space-separated in OAuth requests; individual client controls may use another format. Use that client’s documentation. Do not reuse a copied login or consent URL because its state and PKCE values belong to one login attempt.

Clients needing a pre-registered OAuth application must be coordinated with Tagboard support. The shared metadata URL ending in mcp-inspector has specific callback URLs and is not a universal client ID for arbitrary apps. A client that only supports DCR, static API keys, or legacy SSE is not automatically compatible. Never disable TLS or issuer/audience validation to make a connection work.

Your First Check

Get started by running these prompts in order. They only look at your account, none of them will change anything on air: 

1. Confirm you're connected. Select the Tagboard connection, then ask: "Show my Tagboard profile and available teams. Do not change any production."
2. Find your productions. Pick a team from the list. Ask: "List productions for business unit [your team ID]. Do not change anything." 
3. Look inside one production. Ask: "Show the playlist for [production ID]. Do not preview, take live, or change the production."

Results come 20 at a time (50 at most). So if you get a long list back, you can always ask your AI assistant for the next page.

On multiple Tagboard teams? Here's how your team gets chosen:

• We recommend setting a default team. Just let your AI Assistant know which team should be your default. If you belong to just one team, it's picked and saved for you.
• If you name a team in your request, the assistant uses it for that request only.
• If you don't name a team in your request, the assistant uses your saved default team. 
• Your team access is checked every time. A saved default won't keep working if you lose access to that team. 

What you can ask it to do

Task	What you can ask	Changes anything?
Check identity and teams	Show my profile, list my teams, set my default team	Only the default-team setting
Look at productions	List productions, show a production's state, show its playlist, show what's live, show what's in preview, search playlist items	No
Preview	Put an item in preview, clear preview	Preview only, not on air
Live output	Take an item live, clear live output	Yes, changes what's on air
Playback	Play or pause the playlist, play or pause a video, turn looping on or off, mute or unmute	Yes, can affect on-air output right away

Staying safe when you're live

Live and playback commands can change what viewers see immediately. Treat them the way you'd treat the buttons inside Tagboard Producer.

• Practice on a test production first. Use one that isn't attached to a live broadcast, and confirm the assistant can preview and read state there before you rely on it during a show.
• Read the confirmation before you approve. Check the team, the production, the item, and the action. If any of them is wrong, say no.
• Ask for preview first. For example: "Preview [item title] in [test production] for business unit [team ID]. Show me the exact item you found and ask before changing the preview. Do not take it live."
• "Accepted" does not mean "on screen." When the assistant reports that a command was accepted, it means Tagboard received it. Check the actual display to confirm it rendered.
• If the result is unclear, check before retrying. Ask for the current state first. Sending the same command twice can trigger the action twice.
• Never use a live command as a connection test. Use the read-only checks above.
• Playlist play and pause also affect video. Pausing the playlist pauses the video playing in it.

Troubleshooting

What you see	What to do
The connector or plugin doesn't appear	Check that you're in the right Claude, ChatGPT account and workspace, and that the connector is turned on for your conversation.
No option to add a custom connection	Your plan or workspace may restrict it. Ask your workspace administrator, or use the browser version.
Error on the login page before the permission screen	Start a fresh connection from the assistant. Check the address and that the login is at tagboard.auth0.com. Send support the error details, not the full login link.
Connected, but reads fail	Reconnect and approve the permissions when asked, or contact support.
Reads work, but preview, live, or playback fails	Your connection probably doesn't have that permission yet. Check permissions to confirm.
No permission screen appeared	You may have approved before. Test with a read.
"Business unit selection required"	Ask for your list of teams and choose one.
"Business unit not authorized"	Pick a team from your list, or ask your Tagboard administrator to check your team membership.
"Production not found"	Check the team and production ID. That production may have been deleted so it no longer exists.
"State conflict"	Ask for the current live state, then confirm what you want to do again.
"Rate limited" or "server busy"	Wait a moment and try again. Ask for fewer things at once.
"Guardrail unavailable"	Stop trying to control anything and contact support.
"Command outcome unknown"	Check the state and the actual screen before you retry.
"Result too large"	Ask for a narrower search or smaller pages.
You keep getting asked to sign in again	Reconnect through the connector settings. Don't try to work around it.

Need help getting started? Reach out to your dedicated Client Success Manager or support@tagboard.com.

To help us solve it faster include:

• Which assistant you use and its version (for example, Claude Desktop or ChatGPT in Chrome)
• The date, time, and your time zone
• What you asked and the exact error message
• Your team and production IDs
• A correlation ID, if the error shows one

Before you send screenshots, cover up anything that looks like a sign-in token, cookie, secret, or a long login link. Those should never be shared.