Brightsy MCP Server

This is a Model Context Protocol (MCP) server that connects to an Brightsy AI agent.

Installation

npm install

Usage

To start the server:

npm start -- --agent-id= --api-key=

Or with positional arguments:

npm start --   [tool-name] [message]

You can also provide an initial message to be sent to the agent:

npm start -- --agent-id= --api-key= --message="Hello, agent!"

Customizing the Tool Name

By default, the MCP server registers a tool named "brightsy". You can customize this name using the --tool-name parameter:

npm start -- --agent-id= --api-key= --tool-name=

You can also set the tool name as the third positional argument:

npm start --

Or using the BRIGHTSY_TOOL_NAME environment variable:

export BRIGHTSY_TOOL_NAME=custom-tool-name
npm start -- --agent-id= --api-key=

Environment Variables

The following environment variables can be used to configure the server:

• BRIGHTSY_AGENT_ID: The agent ID to use (alternative to command line argument)
• BRIGHTSY_API_KEY: The API key to use (alternative to command line argument)
• BRIGHTSY_TOOL_NAME: The tool name to register (default: "brightsy")

Testing the agent_proxy Tool

The agent_proxy tool allows you to proxy requests to an Brightsy AI agent. To test this tool, you can use the provided test scripts.

Prerequisites

Before running the tests, set the following environment variables:

export AGENT_ID=your-agent-id
export API_KEY=your-api-key
# Optional: customize the tool name for testing
export TOOL_NAME=custom-tool-name

Alternatively, you can pass these values as command-line arguments:

# Using named arguments
npm run test:cli -- --agent-id=your-agent-id --api-key=your-api-key --tool-name=custom-tool-name

# Using positional arguments
npm run test:cli -- your-agent-id your-api-key custom-tool-name

Running the Tests

To run all tests:

npm test

To run specific tests:

# Test using the command line interface
npm run test:cli

# Test using the direct MCP protocol
npm run test:direct

Test Scripts

1. Command Line Test (test-agent-proxy.ts): Tests the agent_proxy tool by running the MCP server with a test message.

2. Direct MCP Protocol Test (test-direct.ts): Tests the agent_proxy tool by sending a properly formatted MCP request directly to the server.

How the Tool Works

The MCP server registers a tool (named "brightsy" by default) that forwards requests to an OpenAI-compatible AI agent and returns the response. It takes a messages parameter, which is an array of message objects with role and content properties.

Example usage in an MCP client:

// Using the default tool name
const response = await client.callTool("brightsy", {
  messages: [
    {
      role: "user",
      content: "Hello, can you help me with a simple task?"
    }
  ]
});

// Or using a custom tool name if configured
const response = await client.callTool("custom-tool-name", {
  messages: [
    {
      role: "user",
      content: "Hello, can you help me with a simple task?"
    }
  ]
});

The response will contain the agent's reply in the content field.