Actions

In the context of MCP, an Action is the result of an AI agent successfully calling a tool. It represents the "work" being done on behalf of the user.

Lifecycle of an Action

  1. Model Decides: The LLM determines an action is needed based on the user prompt.
  2. Client Invokes: The MCP client sends a call_tool request to the server.
  3. Server Executes: The MCP server performs the operation (e.g., an API call).
  4. Result Returned: The server sends the tool's output back to the model context.

Actions allow agents to interact with the world beyond the confines of their static training data.

Reliable Execution with HasMCP

In a production environment, an Action is more than just a function call—it’s a transaction that requires security and reliability. HasMCP enhances the action lifecycle by providing Secure Secret Management, ensuring that when a client invokes an action, all necessary API keys and tokens are securely injected from an encrypted vault. By combining this with Zero-Trust principles, HasMCP ensures that every action performed by an AI agent is audited, authorized, and executed with the highest level of security.

Questions & Answers

What is the difference between a Tool and an Action in MCP?

In MCP, a Tool is the executable function exposed by a server, whereas an Action is the specific instance of an AI agent invoking that tool to perform work.

Who controls the execution of an Action?

Actions are model-controlled; the AI model decides when to invoke a tool based on the user's prompt and the available tools discovered via the MCP protocol.

What is the typical lifecycle of an MCP Action?

The lifecycle involves the model deciding to act, the client sending a call_tool request, the server executing the operation, and the results being returned to the model's context.

Back to Glossary