Integrating AI Assist with an MCP Server
The instructions below are for desktops and laptops only.
The MCP Server integration is designed to be flexible and secure. You connect tawk.to to an existing MCP-compatible service, choose how the connection is authenticated, and allow AI Assist to use the provided context to improve conversations and automations.
What is an MCP Server?
In simple terms:
- tawk.to requests relevant context for the current chat session.
- The MCP Server supplies structured information and available actions.
- AI Assist uses that information to respond more accurately, guide conversations, and power automations.
This allows AI Assist to work with live, session-level data instead of relying only on predefined responses.
To learn more about the Model Context Protocol itself, see:
What is the Model Context Protocol (MCP)? — official documentation
What is the MCP and how does it work? — Google’s guide
Key benefits of integrating AI Assist with an MCP Server
- Enable smarter automations using data from connected systems.
- Provide AI Assist with up-to-date context during conversations.
- Use secure authentication methods such as API keys or tokens.
- Keep setup simple with minimal configuration.
- Resolve conversations faster and improve agent efficiency.
Before you start
- A remote MCP server that is publicly reachable over HTTPS, and its Remote URL.
- Any authentication your server requires, such as an API key, bearer token, or custom headers.
- A desktop or laptop computer. MCP integration setup is not available on mobile devices.
Steps to set up an MCP Server integration

2. Click Automations on the left navigation bar.

3. Click Integrations in the left submenu.

4. Click MCP Server under Custom Tool.


5. This opens the Overview tab. Click Install in the upper right corner.

6. In the pop-up, enter the following details:
- Server name – A clear name to identify this MCP connection.- Description – Optional details about what the server provides.
- Server URL – The Remote URL for your MCP server.
- Authentication method – Choose API key, bearer token, or custom headers.
Then, click Save.

7. After saving, tawk.to connects to the MCP Server and makes the available context and actions accessible to AI Assist. Once installed, click Agents in the left submenu.

8. Select your AI agent. Then, click Tools in the left menu. Ensure the MCP integration is enabled for your AI agent.

Best practices
Only allow access to the information your AI agent actually needs. Use scoped credentials and authentication headers to control access and protect sensitive data.
Limitations
- Tools are read on connect
AI Assist reads your server's available tools when you connect or refresh the integration. If you add, remove, or change tools, refresh the integration to update the tool list. Changes are not reflected automatically. - Each tool call must complete within ~25 seconds
Requests that take longer will time out. - Interactive and streaming tools are not supported
Tools that pause to ask for input during a run or stream partial progress aren't supported. Each tool runs as a single request and response.
Troubleshooting
If the integration shows an error or Available Tools: 0, check the following:
- Remote URL
Ensure the Remote URL is correct and publicly reachable over HTTPS. - Authentication
Ensure your authentication details are correct. A rejected API key or bearer token shows as an authentication error. - MCP server endpoint
Ensure the endpoint is an MCP server, not a REST API or a static manifest. - Available tools
Ensure the server registers at least one tool. Available Tools: 0 means tawk.to connected successfully but the server exposed no usable tools. - Response time
Ensure the server responds in time. A very slow server may time out. - Recent server changes
If you recently updated the server, refresh the integration to update the tool list.
Related guides
- Click the green live chat icon
- Schedule a call with us
- Visit our community
