MCP Tools Documentation

There are two ways to connect MCP tools to your Gola agents. Pick the one that fits your deployment style.

Example 1: Use What's Already Installed

Got Node and Python on your system? Use them directly with the command approach:

# mcp-agent-inherited.yaml
agent:
  name: "Development Assistant"
  description: "AI assistant with filesystem and time tools"

llm:
  provider: openai
  model: "gpt-4.1-mini"
  auth:
    api_key_env: "OPENAI_API_KEY"

mcp_servers:
  - name: "filesystem"
    command:
      command: "npx"
      args: ["-y", "@modelcontextprotocol/server-filesystem", "."]
    tools: all
    enabled: true
    
  - name: "time"
    command:
      command: "uvx"
      args: ["mcp-server-time"]
    tools: all
    enabled: true

What you need:

• Node.js with npm (for the npx command)
• Python with uv (for the uvx command)

Run it:

# Ensure prerequisites are installed
npm --version  # Check Node.js
uvx --version   # Check uv

# Run the agent
gola --config mcp-agent-inherited.yaml

Example 2: Let Gola Handle Everything

Don't have Node or Python? Don't want to install them? No problem. Use the execution_type approach and Gola downloads what it needs:

# mcp-agent-autodownload.yaml
agent:
  name: "Portable Assistant"
  description: "AI assistant that works anywhere"

llm:
  provider: openai
  model: "gpt-4.1-mini"
  auth:
    api_key_env: "OPENAI_API_KEY"

mcp_servers:
  - name: "filesystem"
    execution_type:
      type: runtime
      runtime: "nodejs"
      entry_point: "@modelcontextprotocol/server-filesystem"
      args: ["."]
    tools: all
    enabled: true
    
  - name: "time"
    execution_type:
      type: runtime
      runtime: "python"
      entry_point: "mcp-server-time"
      args: ["--local-timezone", "UTC"]
    tools: all
    enabled: true

What you need: Nothing. Seriously.

Run it:

# Run it - Gola handles the dependencies
gola --config mcp-agent-autodownload.yaml

What happens behind the scenes:

1. Gola checks if it needs Node.js or Python
2. Downloads them to a local cache if missing
3. Installs the MCP server packages
4. Starts everything with proper isolation

Pro-tip: This approach is perfect for Docker containers or CI/CD pipelines where you don't want to pre-install runtimes.

For details on the auto-download mechanism, see Runtime Auto-Download.

Which Approach Should You Use?

Use the command approach when:

• You're running on your dev machine with tools already installed
• You need specific versions of Node or Python
• You want faster startup (no download overhead)

Use the execution_type approach when:

• You're deploying to environments you don't control
• You want true "zero dependencies" deployment
• You're building Docker containers and want to keep them minimal
• Different team members have different setups

Both approaches use the exact same MCP servers - it's just about how they get launched.