This is the multi-page printable view of this section. Click here to print.
Tutorials are learning-oriented guides that take you through a series of steps to complete a project. They focus on learning by doing, and helping beginners get started with the system.
These tutorials will help you get familiar with gomcptest and its components.
This tutorial will guide you through setting up the gomcptest system and configuring Google Cloud authentication for the project.
gomcptest is a proof of concept (POC) implementation of the Model Context Protocol (MCP) with a custom-built host. It enables AI models like Google’s Gemini to interact with their environment through a set of tools, creating powerful agentic systems.
The project consists of three main parts:
Host Components:
MCP Tools:
MCP Protocol: The standardized communication layer that allows models to discover, invoke, and receive results from tools
gomcptest enables a variety of agent-based applications:
Before using gomcptest with Google Cloud Platform services like Vertex AI, you need to set up your authentication.
If you haven’t already configured the Google Cloud CLI, run:
gcloud init
This interactive command will guide you through:
Authenticate your gcloud CLI with your Google account:
gcloud auth login
This will open a browser window where you can sign in to your Google account.
Application Default Credentials are used by client libraries to automatically find credentials when connecting to Google Cloud services:
gcloud auth application-default login
This command will:
~/.config/gcloud/application_default_credentials.json)These credentials will be used by gomcptest when interacting with Google Cloud services.
Clone the repository:
git clone https://github.com/owulveryck/gomcptest.git
cd gomcptest
Build Tools: Compile all MCP-compatible tools
make tools
Choose Interface:
After completing the basic setup:
tools directoryThis tutorial will guide you step-by-step through running and configuring the OpenAI-compatible server in gomcptest. By the end, you’ll have a working server that can communicate with LLM models and execute MCP tools.
gcloud auth loginThe OpenAI server requires several environment variables. Create a .envrc file in the host/openaiserver directory:
cd host/openaiserver
touch .envrc
Add the following content to the .envrc file, adjusting the values according to your setup:
# Server configuration
PORT=8080
LOG_LEVEL=INFO
IMAGE_DIR=/tmp/images
# GCP configuration
GCP_PROJECT=your-gcp-project-id
GCP_REGION=us-central1
GEMINI_MODELS=gemini-2.0-flash
IMAGEN_MODELS=imagen-3.0-generate-002
Ensure the image directory exists:
mkdir -p /tmp/images
Load the environment variables:
source .envrc
Now you can start the OpenAI-compatible server:
cd host/openaiserver
go run . -mcpservers "../bin/GlobTool;../bin/GrepTool;../bin/LS;../bin/View;../bin/Bash;../bin/Replace"
You should see output indicating that the server has started and registered the MCP tools.
Open a new terminal window and use curl to test the server:
curl http://localhost:8080/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{
"model": "gemini-2.0-flash",
"messages": [
{
"role": "user",
"content": "Hello, what can you do?"
}
]
}'
You should receive a response from the model explaining its capabilities.
Now let’s test function calling by asking the model to list files in a directory:
curl http://localhost:8080/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{
"model": "gemini-2.0-flash",
"messages": [
{
"role": "user",
"content": "List the files in the current directory"
}
]
}'
The model should respond by calling the LS tool and returning the results.
In this tutorial, you’ve:
Now that you have a working OpenAI-compatible server, you can:
Check out the How to Configure the OpenAI Server guide for more advanced configuration options.
This tutorial guides you through setting up and using the cliGCP command line interface to interact with LLMs and MCP tools. By the end, you’ll be able to run the CLI and perform basic tasks with it.
gcloud auth loginThe cliGCP tool is a command-line interface similar to tools like Claude Code. It connects directly to the Google Cloud Platform’s Vertex AI API to access Gemini models and can use local MCP tools to perform actions on your system.
First, build the cliGCP tool if you haven’t already:
cd gomcptest
make all # This builds all tools including cliGCP
If you only want to build cliGCP, you can run:
cd host/cliGCP/cmd
go build -o ../../../bin/cliGCP
The cliGCP tool requires environment variables for GCP configuration. You can set these directly or create an .envrc file:
cd bin
touch .envrc
Add the following content to the .envrc file:
export GCP_PROJECT=your-gcp-project-id
export GCP_REGION=us-central1
export GEMINI_MODELS=gemini-2.0-flash
export IMAGEN_MODELS=imagen-3.0-generate-002
export IMAGE_DIR=/tmp/images
Load the environment variables:
source .envrc
Now you can run the cliGCP tool with MCP tools:
cd bin
./cliGCP -mcpservers "./GlobTool;./GrepTool;./LS;./View;./dispatch_agent -glob-path ./GlobTool -grep-path ./GrepTool -ls-path ./LS -view-path ./View;./Bash;./Replace"
You should see a welcome message and a prompt where you can start interacting with the CLI.
Let’s try a few simple interactions:
> Hello, who are you?
You should get a response introducing the agent.
Now let’s try using some of the MCP tools:
> List the files in the current directory
The CLI should call the LS tool and show you the files in the current directory.
> Search for files with "go" in the name
The CLI will use the GlobTool to find files matching that pattern.
> Read the README.md file
The CLI will use the View tool to show you the contents of the README.md file.
Let’s create a simple task that combines multiple tools:
> Create a new file called test.txt with the text "Hello, world!" and then verify it exists
The CLI should:
In this tutorial, you’ve:
Now that you’re familiar with the cliGCP tool, you can:
Check out the How to Configure the cliGCP Tool guide for advanced configuration options.