Skip to main content
App MCP lets external AI assistants like Claude, ChatGPT, and Cursor connect to your published app through its own Model Context Protocol (MCP) server. Connect an AI assistant to view or manage app data, message agents, or run custom tools, based on the access you allow.
Sample requests people can make to a Base44 app from an AI assistant, such as which leads have not been contacted in a week, what is in stock, and adding a new patient

​
What you can do with App MCP

App MCP lets you and other people work with your published app directly from an AI assistant such as Claude, ChatGPT, or Cursor. What an assistant can do depends on your server’s access mode:
  • With Public · No sign-in access, assistants can view records from the entities you make available. They cannot change records, message agents, or run custom tools.
  • With Sign in required · OAuth access, you and other signed-in users can view or change records, message agents, and run custom tools, based on each person’s app permissions.
For example, with OAuth access, you could ask your AI assistant to find leads that need follow-up and add a new contact without opening your CRM app. Other people with the required app permissions could work with the CRM from their own assistants.

​
How App MCP works

Set up MCP access for AI assistants from your app dashboard. Once you publish your app, connected assistants can work with the entities, agents, and custom tools you make available. App MCP controls access at three levels:
  • Access mode: Determines who can connect and whether they need to sign in.
  • Tool access: Determines which entities, agents, and custom tools connected assistants can use.
  • App permissions: Determine what each signed-in user can view or change through an assistant.
The Access badge on the Overview card shows your server’s current access mode. To change it, ask the AI chat.

Looking for a different MCP connection?

App MCP connects external AI assistants to one published app. For other types of MCP connections:

​
Setting up MCP access

Set up App MCP for each app from the app’s MCP page.
The MCP page in the app dashboard with the Set up access button

The MCP page when your app is ready to set up access

If MCP is turned off, reach out to your workspace owner or admin. Learn more about managing MCP access for your workspace.
To set up App MCP access:
  1. Go to your app’s dashboard.
  2. Click MCP.
  3. Click Set up access. This sends a prompt to the AI chat. If you see Enable custom authentication instead, click it and set up your app’s custom login pages. Then return to the MCP page and click Set up access.
  4. Choose Public or Sign in required · OAuth when prompted.
  5. On the MCP page, review the Access mode, Connection URL, and Tool access.
  6. Click Publish to publish your app and activate the server.
App MCP uses the published version of your app:
  • Publish your app before connecting an assistant for the first time.
  • Publish again after changing the access mode or tool access.
  • After you publish tool access changes, each connected user must reconnect their assistant and approve the updated permissions.

​
Managing tool access

Choose which entities, agents, and custom tools connected assistants can use. To manage tool access:
  1. Go to your app’s dashboard.
  2. Click MCP.
  3. In the Tool access section, click Edit.
  4. Choose what assistants can access:
    • Entities: Turn an entity on or off. For enabled entities, choose Full access to allow Read, Create, Update, and Delete actions, or choose Custom access to select specific permissions.
    • Agents: Set each agent to Enabled or Disabled.
    • Custom tools: Turn on each custom tool you want assistants to use.
  5. Click Save changes.
  6. Click Publish to activate your changes.

​
Connecting an AI assistant

Connect your AI client to your app’s MCP server using the connection URL from the MCP page. To connect an AI assistant:
  1. Go to your app’s dashboard.
  2. Click MCP.
  3. On the Overview card, click Copy URL next to Connection URL.
  4. Open the section for your AI assistant below and follow its setup steps.
  1. Add the connector in one click. Click Add connector, and Claude opens with the connector’s details pre-filled.
  2. Review the pre-filled details. If the connection URL is missing, paste it manually.
  3. Click Add and start using it in chat.
  1. Open Settings and choose Plugins in the ChatGPT desktop app. On the web, custom MCP servers go through Developer mode instead.
  2. Open the MCPs tab and click Add server.
  3. Name the server, set Type to Streamable HTTP, and paste the connection URL.
  4. Click Save.
  5. Click Restart, then enable the server. The app restarts to load it, and the server’s tools become available in chat.
  1. Open Cursor Settings, then Tools & Integrations.
  2. Click New MCP Server. Cursor opens your mcp.json configuration.
  3. Add a server entry with the connection URL. Set the entry’s url field to the connection URL and save.
  4. Toggle the server on.
  1. Copy the connection URL.
  2. Add it to your client as a streamable HTTP MCP server. Most clients ask for a name and a URL, with no extra configuration.
  3. Reload the client and start using the tools.
If your server uses Sign in required · OAuth, the client opens your app’s consent page. Sign in with your app account and approve access.

​
Adding in-app connection guidance

After setting up App MCP, you can add a page to your app that shows people how to connect their own AI assistants. They use the instructions to configure their external AI client. To add in-app connection guidance:
  1. Go to your app’s dashboard.
  2. Click MCP.
  3. On the How to connect AI assistant card, click the Send to chat icon next to Ask Base to add in-app guidance to help your users connect AI assistants.
  4. Wait for the AI chat to finish, then click Publish.
Base44 adds a /connect page with instructions for Claude, ChatGPT, Cursor, and custom AI clients.

​
Managing MCP access for your workspace

The workspace MCP policy controls whether apps in the workspace can use App MCP. It applies to every app and overrides app-level settings.
Only workspace owners and admins can update the MCP policy. Policy changes take effect immediately and do not require publishing.
To set the workspace MCP policy:
  1. Open your workspace Settings.
  2. Click Governance.
  3. Under MCP access for AI assistants, select a policy:
    • Allowed: Apps can use App MCP, with or without sign-in.
    • Sign-in required: Apps can use App MCP, but assistants must sign in. Every App MCP server in the workspace uses OAuth.
    • Not allowed: Apps in the workspace cannot use App MCP.
The workspace Governance page with the MCP access for AI assistants drop-down open, showing the Allowed, Sign-in required, and Not allowed policies

The MCP access for AI assistants policy on the Governance page

If you select Not allowed, active App MCP servers stop working on their next request.

​
Troubleshooting

Your workspace policy is blocking App MCP. The notice tells you whether to ask an admin or to change it yourself, and owners and admins can click Open workspace settings.
Your app requires sign-in, but the server is set to Public, so it will not run. On the MCP page you see Your app requires login but your MCP server is exposed publicly. Ask in the AI chat to switch the server to OAuth, or make the app public, then publish.
Your Claude organization may restrict custom connectors. Ask your Claude admin whether custom connectors are enabled for your account.
After you publish changes to the available tools, reconnect the app from the AI client and approve the updated access.