MCP Client

This document describes the MCP client, which consists of a reusable library and a CLI.

Library

The core of the MCP client is a reusable library located in mcp/client/lib. This library provides a simple way to connect to an MCP server, discover its tools, and process queries using the Gemini API.

Usage

To use the library, import the MCPClient class and create a new instance. Then, call the connectToServers method with the server configurations.

import { MCPClient } from './lib/mcp-client';
import { readSettings } from './lib/settings';

const client = new MCPClient('YOUR_API_KEY_HERE');
const settings = await readSettings();
await client.connectToServers(settings.mcpServers);
const response = await client.processQuery('What is the weather like in New York?');
console.log(response);

CLI

The CLI provides a simple REPL to interact with the MCP.

Prerequisites

Node.js and npm installed.
A Google Gemini API Key.

Setup

Install Dependencies: From the root of the repository, run the following command to install the required Node.js packages:
```
npm install
```
Configure API Key: Create a .env file in the root of the repository. Add your Gemini API key to this file:
```
GEMINI_API_KEY="YOUR_API_KEY_HERE"
```
The client uses this file to load your API key securely.
Configure Servers: The client reads a list of MCP servers to connect to from ~/.hades/settings.json. Create this file with the following format:
```
{
  "mcpServers": {
    "weather": {
      "command": "node",
      "args": ["_bazel_bin/mcp/services/weather/index.js"]
    },
    "mainServer": {
      "command": "bin/mcp_server.py"
    },
    "anotherServer": {
      "url": "http://another-mcp-server.com"
    }
  }
}
```
The mcpServers object contains a map of server names to server configurations. Each server configuration can have one of the following forms:
- A url for a remote server.
- A command for a local server.
- A command and args for a local server with arguments.

Building

The client and any services it connects to must be compiled using Bazel.

Build the Client:

bazelisk build //mcp/client/cli:main_ts_lib

Build the Weather Service: To test the client, build the example weather service:
```
bazelisk build //mcp/services/weather:index_ts_lib
```
These commands compile the TypeScript code and place the output in the _bazel_bin directory.

Running the Client

To run the client, you execute its compiled script with node.

Start the Client: Use the following command from the repository root:
```
node _bazel_bin/mcp/client/cli/main.js
```
The client will prompt you to select a server if multiple servers are configured in your settings.json file.
Interact with the Client: Once running, the client will display a list of tools available from the connected service. You can then type queries for the model to answer. For example:
```
Query: What is the weather like in New York?
```
The client will use the Gemini API and the connected service's tools to respond. To exit, type quit.
Note: The get-forecast tool requires latitude and longitude as input. For a more seamless experience, you can connect the client to a search MCP server that can provide these coordinates for a given location.