Eino

You can extend your AI agent with Model Context Protocol (MCP) to discover and run tools during conversations. This server setup lets your agent query real-time information, invoke MCP tools, and interact through a streaming or stdio-based transport, enabling flexible workflows between your model and practical utilities.

How to use

After you start the MCP-enabled agent and its tools, you interact through a simple command interface. You can ask it to fetch the current time, or ask it to perform tasks using the MCP tools it discovers. The system supports interactive prompts and continuous sessions, so your agent can chain tool calls as part of a conversation. To exit, type exit or bye.

Key usage patterns include: querying real-time information via the Time Tool, leveraging the tool registry to discover capabilities, and passing context between the model and tools through MCP. The terminal-like interface responds with formatted results from the tools and model, creating an AI assistant that can reason with up-to-date data.

How to install

Prerequisites you need before starting: Go 1.23.4 or higher. Verify your Go version with the following command.

go version

Prepare your environment by creating and populating an environment configuration file. Copy the template and fill in your API credentials and endpoints.

cp env.sh.example env.sh

# Edit env.sh to set your credentials
cat env.sh

export OPENAI_API_URL="https://dashscope.aliyuncs.com/compatible-mode/v1"
export MODEL_ID="qwen2.5-32b-instruct"
export OPENAI_API_KEY="your-api-key-here"

Start the MCP time tool servers in separate terminals to provide time-related capabilities to MCP. You can choose either the mcp-go implementation or the go-mcp implementation.

Starting the MCP time tools

Option 1: using the mcp-go implementation. Build and run the server with SSE transport and a local listen address.

cd tools/mcp-time
go build -o mcp-time main.go
./mcp-time -transport=sse -server_listen=localhost:8080

Option 2: using the go-mcp implementation. Build and run the server with SSE transport.

cd tools/mcp-time-v2
go build -o mcp-time-v2 main.go
./mcp-time-v2 -transport=sse

Start the main application

With the MCP time tools running, start the main AI application that uses MCP to access those tools.

# From the project root, start the main app
go run main.go

Additional notes

Environment variables to configure the MCP-enabled setup include API details and model settings. Ensure you provide valid values in env.sh for your OpenAI API URL, model ID, and API key. The time tools communicate over the SSE transport, so ensure port 8080 (or your chosen port) is accessible for the MCP client.

Troubleshooting and tips

• If you see transport-related errors, verify you started the tool server with -transport=sse and that the client connects to the correct SSE endpoint.
• Check that the OpenAI API URL, model ID, and API key are correctly set in env.sh.
• Ensure the MCP tools have started before launching the main application to avoid initialization failures.
• Use exit or bye to gracefully terminate the interactive session.

Available tools

Time Tool

MCP tool that provides current time information, supports time zones, and returns a formatted time string.