Core Concepts

Actions

Actions are the connection between your AI agents and your backend APIs. They enable agents to perform operations like fetching data, creating orders, or updating records by calling your existing endpoints.

What are Actions?

In Meebly, an "action" is usually your backend API endpoint that agents can call. When you create an action in the Meebly dashboard, you define:

The operation name (what the action does)
HTTP method (GET, POST, PUT, DELETE, PATCH)
API path with parameter placeholders
Input fields and their types
System instructions for using the action
Whether human approval is required (HITL)

Actions are created in the Meebly dashboard and automatically become available to agents in the same environment. You don't need to write integration code - agents learn how to use your actions from the configuration.

Action Configuration

Each action is configured with these properties:

Property	Description	Example
`operationName`	Name of the operation (used by AI)	getUserProfile, createOrder
`method`	HTTP method	GET, POST, PUT, DELETE, PATCH
`path`	API endpoint path	/api/users/{userId}
`systemInstructions`	Context for when and how to use this action	"Use this to fetch user profile data"
`requestConfirmation`	Require human approval before execution	true, false
`inputFields`	Parameter definitions (name, type, description)	[{name: "userId", type: "string"}]

How Agents Use Actions

When a user makes a request, the agent:

Analyzes the user's intent
Identifies relevant actions via semantic filtering
Determines required parameters from context
Requests approval if action has requestConfirmation=true
Executes the API call with your backend token
Processes the response and continues the conversation

json

// Example action call in SSE stream
{
  "type": "meebly-event",
  "event": {
    "type": "tool-call",
    "payload": {
      "toolCallId": "call_abc123",
      "toolName": "getUserProfile",
      "args": {
        "userId": "user_456"
      }
    }
  }
}

Testing Actions

To test an action in isolation, you can use the Sandbox page to chat with an agent that only has access to the new action you've created. This is ueful for verifying that the action is configured correctly before you add it to one of your agents.

Semantic Filtering

Agents don't call every action on every request. Meebly uses semantic filtering to automatically select only the relevant actions based on:

The user's question or request
Action names and system instructions
Current conversation context
Screen state (if using Screen Observer)

Pro Tip: Write descriptive operation names and clear system instructions. Instead of "endpoint1", use "getUserOrderHistory" with instructions like "Use this to fetch a user's past orders when they ask about order history or want to reorder items."

PreviousEnvironments NextMCP Servers

Last updated: March 2026Report an issue