ollama

You connect local Ollama models to MCP-compatible applications with this MCP server. It enables structured task decomposition, result evaluation, model management, and standardized communication, helping you build robust AI-powered workflows that split problems into manageable steps and assess outcomes.

How to use

You interact with the MCP server through an MCP client to perform tasks such as decomposing complex work, evaluating results, and running Ollama models. Start tasks, request subtask breakdowns, and feed results back for evaluation. Use the server to manage conversation with local Ollama models and to route results through standardized MCP prompts and tools.

Typical usage patterns include: decomposing a large task into subtasks, running a model on a given prompt, and then evaluating the resulting outputs against defined criteria. You can chain these steps to build end-to-end workflows where each stage feeds into the next, with clear metadata and structured error reporting to quickly identify where things go wrong.

How to install

pip install ollama-mcp-server

Install prerequisites: Python 3.8+ and a working Python environment. You also need Ollama installed and running locally to access models.

Install Ollama and prepare models locally before starting the MCP server.

## Additional sections

Configuration and runtime behavior are designed to be practical and transparent. You can tune performance and error handling to suit your environment, and you can see detailed error messages that help you diagnose issues quickly.

Environment variables you may set include the host, default model, and log level. Basic examples show how to point the server at your Ollama instance and set a preferred model.

Troubleshooting tips include checking that Ollama is running, ensuring the configured model exists, and reviewing the detailed error structure that the server returns when something goes wrong.

## Available tools

### add-task

Create a new task with a name and description. Optional fields include priority, deadline, and tags. Returns the task identifier.

### decompose-task

Break down a complex task into subtasks using a specified granularity. Optionally limit the number of subtasks.

### evaluate-result

Evaluate a given result against specified criteria and provide feedback, including a detailed score and improvement suggestions.

### run-model

Execute an Ollama model with a provided prompt and optional parameters such as temperature and max\_tokens.