MCP: what they are and how to use them

Model Context Protocol (MCP) is an open standard for communication between AI Agents and external systems.

The platform includes many pre-configured MCPs, but you can also create your own.

MCPs enable two-way communication between LLMs and external resources, letting your Agent interact easily with all the tools (tools) made available by the MCP server.

MCP servers: what are they and how do they work under the hood?

What MCPs are

MCP servers expose "capabilities" that your AIsuru Agent can use. Among these "capabilities" you'll find:

• Tools -> actions your Agent can perform (e.g., to create a calendar event);

• Resources -> data your Agent can read (e.g., today's events);

• Prompts -> predefined instruction templates provided by the MCP server — shortcuts for the most common interactions.

How MCPs work

When you add an MCP to your Agent on AIsuru, the platform automatically creates two distinct advanced functions:

• One lets it read the list of available tools;

• The other lets it execute those tools.

Whenever the LLM decides it's appropriate, your Agent will:

• Call the function that reads the list of available tools;

• Decide which tools to call and with which parameters (based on the user's request).

How to activate and test an MCP

MCPs on AIsuru: how to activate them

To activate an MCP:

1. Go to the MCP section of your Agent:

   1. From the AIsuru homepage, press "Manage" next to the Agent;

   2. In the sidebar, press "MCP".

2. In the MCP grid you can:

   1. Create a custom MCP;

   2. Activate an existing one.

For each MCP you'll find:

• The MCP name (e.g., "MongoDB MCP Server", "PostgreSQL MCP Server");

• A category tag (e.g., "mongodb", "generic");

• A short description;

• The number of parameters for the MCP.

What to do after activating an MCP

After activating your MCP, it's essential to:

1. Update your Agent's instructions: explain what the MCP is for and when and how to use it;

2. Test the MCP: to test it, open a conversation with your Agent (use the "Try" section). If something's not working, check the conversation history to see the MCP's response. If the MCP isn't being called, make sure you've explained more clearly in the instructions when it should be called, or explicitly ask the Agent to use the MCP to answer your question.

If you have any doubts, you can ask Manuela for help or open a support ticket.