gollama-mcp-server

An MCP (Model Context Protocol) server written in Golang to interact with the Ollama API. This server provides tools for AI assistants to interact with local Ollama models through the MCP protocol.

Features

This MCP server provides the following tools for interacting with Ollama:

• listModels - List all available models in your Ollama instance
• pullModel - Download models from the Ollama library
• pushModel - Push models to a registry
• deleteModel - Remove models from your local instance
• showModel - View detailed model information including modelfile, parameters, and template
• copyModel - Create a copy of a model with a different name
• embeddings - Generate vector embeddings for text

Prerequisites

• Go 1.24.4 or later
• Ollama installed and running locally (or accessible via network)

Configuration

The server reads the Ollama host from the OLLAMA_HOST environment variable. If not set, it defaults to http://localhost:11434.

Example values:

export OLLAMA_HOST=http://localhost:11434
export OLLAMA_HOST=http://192.168.0.185:11434
export OLLAMA_HOST=http://spark.lan:11434

Installation

# Clone the repository
git clone <repository-url>
cd gollama-mcp-server

# Build the binary
go build -o bin/gollama-mcp-server ./cmd/main.go

Usage

STDIO Mode (for MCP clients)

Run the server in STDIO mode for use with MCP-compatible clients:

./bin/gollama-mcp-server

HTTP Mode (Streamable HTTP)

Run the server with HTTP transport on a specific port:

./bin/gollama-mcp-server -port 8080

You can also specify a custom host:

./bin/gollama-mcp-server -host 127.0.0.1 -port 8080

Docker

Using Pre-built Image

A pre-built Docker image is available from GitHub Container Registry:

docker pull ghcr.io/kevensen/gollama-mcp-server:latest

Building the Docker Image

Alternatively, build the Docker image locally from the repository root:

docker build -t gollama-mcp-server .

Running with Docker

Run the server in HTTP mode (port 8080) using the pre-built image:

docker run -p 8080:8080 ghcr.io/kevensen/gollama-mcp-server:latest

Or use your locally built image:

docker run -p 8080:8080 gollama-mcp-server

If your Ollama instance is running on the host machine, use host networking:

# Linux
docker run --network host ghcr.io/kevensen/gollama-mcp-server:latest

# macOS/Windows - use host.docker.internal
docker run -p 8080:8080 -e OLLAMA_HOST=http://host.docker.internal:11434 ghcr.io/kevensen/gollama-mcp-server:latest

To connect to a remote Ollama instance:

docker run -p 8080:8080 -e OLLAMA_HOST=http://your-ollama-host:11434 ghcr.io/kevensen/gollama-mcp-server:latest

Run in detached mode with automatic restart:

docker run -d --restart unless-stopped \
  -p 8080:8080 \
  -e OLLAMA_HOST=http://host.docker.internal:11434 \
  --name gollama-mcp \
  ghcr.io/kevensen/gollama-mcp-server:latest

Tool Usage Examples

List Models

{
  "name": "listModels"
}

Generate Completion

{
  "name": "generate",
  "arguments": {
    "model": "llama3.2",
    "prompt": "Write a haiku about programming",
    "temperature": 0.7,
    "stream": false
  }
}

Chat

{
  "name": "chat",
  "arguments": {
    "model": "llama3.2",
    "messages": [
      {"role": "user", "content": "Hello! How are you?"}
    ],
    "temperature": 0.8,
    "stream": false
  }
}

Pull Model

{
  "name": "pullModel",
  "arguments": {
    "model": "llama3.2",
    "stream": true
  }
}

Generate Embeddings

{
  "name": "embeddings",
  "arguments": {
    "model": "nomic-embed-text",
    "input": "The quick brown fox jumps over the lazy dog"
  }
}

Development

Project Structure

gollama-mcp-server/
├── cmd/
│   └── main.go              # Application entry point
├── internal/
│

... [View full README on GitHub](https://github.com/kevensen/gollama-mcp-server#readme)