Skip to main content
In Universal MCP, Applications are the fundamental building blocks for connecting your AI agents to external services and APIs. They act as specialized connectors or drivers that understand how to communicate with a particular service, manage its authentication, and expose its functionalities as usable Actions (Tools).

​
The Role of an Application

An Application in Universal MCP is responsible for:
  1. Connection Management: Establishing and maintaining communication with the external service. This could involve setting base URLs for HTTP APIs or handling GraphQL client instances.
  2. Authentication: Securely managing and applying the necessary credentials (API keys, OAuth tokens, etc.) to authenticate requests to the service. This is often handled in conjunction with an Integration.
  3. Defining Operations (Actions/Tools): Exposing the capabilities of the external service as a list of well-defined functions (Actions or Tools) that an AI agent can call.
  4. Data Transformation: Potentially transforming data to and from the format expected by the external service and the format understood by the AI agent.
  5. Error Handling: Managing and standardizing errors that might occur during interaction with the external service.

​
Base Application Classes

Universal MCP provides several base classes to simplify the development of new applications, catering to common communication patterns:
  • BaseApplication: This is the abstract base class that all applications must inherit from. It defines the common interface, including the essential list_tools() method which every application must implement to declare the tools it provides.
  • APIApplication: This class extends BaseApplication and is designed for applications that interact with standard HTTP/RESTful APIs. It includes:
    • An integrated httpx.Client for making HTTP requests (_get, _post, _put, _delete, _patch).
    • Logic to automatically include authentication headers based on the configured Integration.
    • Management of a base_url for the API.
  • GraphQLApplication: A specialized class extending BaseApplication for services that use GraphQL APIs. It provides:
    • An integrated gql.Client for executing GraphQL queries and mutations.
    • Similar to APIApplication, it handles authentication headers derived from its Integration.
    • Methods like query() and mutate() for interacting with the GraphQL endpoint.

​
Creating a New Application

When you need to connect to a new service not yet supported, you’ll typically create a new Python class that inherits from one of these base classes. For example, if you’re integrating a new REST API, you’d likely inherit from APIApplication. The key aspects you’ll implement are:
  • The constructor (__init__) to set up any specific configurations (like the base_url).
  • Methods that represent the individual tools (Actions) the application offers.
  • The list_tools() method to return a list of these tool methods.
By encapsulating the specifics of interacting with an external service within an Application, Universal MCP promotes modularity and reusability. Your AI agents can then interact with a variety of services through a consistent tool-based interface. Refer to applications/application.py for more information.