Ollama MCP Server

Ollama MCP Server is an MCP server that integrates local Ollama large language model instances with Model Context Protocol applications. Built for AI developers, system engineers, and local AI workflow creators, this server exposes local Ollama endpoints directly to host clients like Claude Desktop. It enables clients to execute queries against local open-source models such as Llama 3, Mistral, or Qwen2, allowing users to run inference without sending data to external APIs. In addition to direct text generation, the server coordinates structured workflows including multi-step task decomposition and result validation against explicit evaluation criteria. It structures interactions through standardized URI resources covering tasks, results, and available models. The implementation features HTTP connection pooling and an LRU cache to reduce latency and Ollama server resource usage, alongside detailed error handling that returns helpful context such as lists of valid models when a requested model is missing.

Category: AI & LLM Tooling

Tags: inference, llm, local-llm, ollama

Visit Ollama MCP Server

How to install and configure Ollama MCP Server

1. Ensure Ollama is installed and running locally, then pull your desired model (e.g., ollama pull llama3). 2. Install the server package using pip: bash pip install ollama-mcp-server 3. Configure Claude Desktop by editing claude_desktop_config.json (located at ~/Library/Application Support/Claude/claude_desktop_config.json on macOS or %APPDATA%/Claude/claude_desktop_config.json on Windows). 4. Add the server configuration under the mcpServers section using uvx: json { "mcpServers": { "ollama-MCP-server": { "command": "uvx", "args": ["ollama-MCP-server"] } } } 5. Restart Claude Desktop to load the tools.

What you can do with Ollama MCP Server

• Running local LLM inference directly from Claude Desktop using installed Ollama models like Llama 3 or Mistral. - Decomposing complex engineering challenges into manageable subtasks categorized by granularity levels, dependencies, and estimated complexity. - Evaluating generated task outputs against structured validation criteria including accuracy, completeness, and clarity scores. - Managing local model inventory and inspecting available Ollama model endpoints via standardized model URI resources. - Caching repeated model completions using built-in LRU caching and HTTP connection pooling to minimize local hardware load.

Key facts

• Open Source
• https://github.com/NewAITees/ollama-MCP-server
• AI & LLM Tooling, Developer Tools & Code Intelligence
• inference, llm, local-llm, ollama

Part of MCP Servers

What is Ollama MCP Server?

Ollama MCP Server is an open-source Model Context Protocol server that bridges local Ollama LLM deployments with MCP-compatible clients. It gives client assistants the ability to execute model queries, break down complex task lists, and evaluate results locally.

What tools are included in Ollama MCP Server?

The server includes add-task for creating tracked tasks, decompose-task for dividing complex problems into subtasks, evaluate-result for analyzing outcomes against specific score criteria, and run-model for running prompts against local Ollama models.

Which MCP clients work with Ollama MCP Server?

Ollama MCP Server works with any client supporting the Model Context Protocol, including Claude Desktop on macOS and Windows, as well as debugging environments like the Model Context Protocol Inspector.

How does the server determine which Ollama model to use?

The server determines the active model using a strict priority order. It first checks the model parameter passed during a tool call, followed by the environment settings in your MCP configuration, the OLLAMA_DEFAULT_MODEL environment variable, and defaults to llama3.

Is Ollama MCP Server open source?

Yes, Ollama MCP Server is open source and distributed under the MIT license on GitHub.