Monite MCP Server

This project provides a Monite MCP (Model Context Protocol) server that seamlessly integrates Monite's advanced AI functionalities as tools for AI agents and AI-driven applications. It offers a streamlined solution for Monite's onboarded clients to leverage AI-driven insights and automation with their financial data, enhancing operational efficiency and decision-making capabilities.

This server is specifically designed to be valuable and easily usable for Monite customers, allowing them to extend their Monite platform experience with powerful AI capabilities.

Status: The MCP server is ready for use but is under active development. New tools and features may be added, and existing ones might be refined.

Main Goals

• Provides a set of tools corresponding to various Monite AI features, exposed via as a MCP server.
• Enables easy integration with AI agent frameworks that support the MCP specification.

Prerequisites

The primary requirement for running the Monite MCP server is a Monite Authentication Service. This service is responsible for providing Entity User authentication tokens, which the MCP server uses to make authorized calls to the Monite AI API.

A Python example implementation of such an authentication service (auth_service.py) is available in the code_examples/ directory.

For comprehensive details on Monite's authorization levels and credential management, please refer to the Monite API Documentation.

Getting Started

This guide assumes you are familiar with the Monite API, possess partner credentials (CLIENT_ID and CLIENT_SECRET), and have an onboarded entity with existing data.

The Monite MCP server is intended for use with Monite Entity User tokens. Consequently, monite-mcp relies on an external authorization service that securely manages your CLIENT_ID and CLIENT_SECRET.

Environment Variables

A Monite MCP server instance requires the following environment variables to be configured:

• AUTH_SERVICE_URL: The URL of your authentication service. This service must expose a GET /token endpoint that accepts an entity_user_id query parameter and returns a Monite Entity User token.
• ENTITY_USER_ID: The ID of the Entity User for whom the MCP server will operate.
• MONITE_AI_API_BASE: (Optional) The base URL for the Monite AI API. Defaults to the Monite Sandbox environment: https://api.sandbox.monite.com/v1/mcp.

The AUTH_SERVICE_URL and ENTITY_USER_ID variables are mandatory for each Monite MCP server instance. Your CLIENT_ID and CLIENT_SECRET will be required by the authentication service itself (see example below).

Running the Example Authentication Service

The code_examples/ directory includes an example authentication service (auth_service.py). To run it:

1. Ensure your CLIENT_ID and CLIENT_SECRET are set as environment variables.
2. Navigate to the code_examples/ directory.
3. Execute the following command:

   make run-auth-service
   

   This will start auth_service.py, which the MCP server can then use to fetch tokens. By default, it runs on http://localhost:8888, so your AUTH_SERVICE_URL would be http://localhost:8888/token.

Running the Monite MCP Server and Example Agents

The Monite MCP server communicates via standard input/output (stdio) and is typically started as a subprocess by an AI agent.

The code_examples/ directory contains pydantic_ai_agent.py and mcp_use_agent.py, which demonstrate how to instantiate and use the monite-mcp server with different agent frameworks.

These examples require OPENAI_API_KEY in the environment :)

To run these examples (ensure the example authentication service is running first):

1. Navigate to the code_examples/ directory.
2. Use one of the following commands:
   • For the Pydantic AI example:

     make run-pydantic-ai-agent
     

   • For the MCP-Use example:

     make run-mcp-use-agent
     

These commands will execute the respective agent scripts, which in turn start and interact with the monite-mcp server.

Active Development

This project is actively being developed. Contributions, bug reports, and feature requests are welcome!