# Ollama MCP Server [Health: Active]

**Category:** 🗣️ Conversational AI  
**Repository:** https://github.com/m-mehdi14/ollama-mcp-server  
**GitHub Stars:** 3  
**npm Downloads (last month):** 436  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/muhammadmehdi-ollama-mcp-server

## Description
Exposes local Ollama models to MCP clients such as Cursor, Claude Desktop and Cline. Local stdio server.

## Tools
Capabilities this server exposes over MCP:

- **name**
- **size**
- **modified_at**
- **digest**

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "ollama-mcp-server": {
    "command": "npx",
    "args": ["-y","@muhammadmehdi/ollama-mcp-server"],
    "env": {
      "OLLAMA_BASE_URL": ""
    }
  }
}
```

**Requires environment variables:** `OLLAMA_BASE_URL` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What Ollama MCP Server does

Ollama MCP Server MCP server provides an MCP interface to an Ollama service running on the same machine. Its listed capabilities expose model metadata: `name`, `size`, `modified_at`, and `digest`. The server is intended for MCP clients that need to query local Ollama model information without connecting to a hosted model service.

The repository identifies Cursor, Claude Desktop, Cline, and other MCP-compatible applications as supported clients. Because the server uses the standard MCP protocol, client compatibility depends on the client being able to launch a local stdio MCP process.

## How it works

The server runs as a local Node.js process and communicates with its MCP client over standard input and output. It then connects to the local Ollama HTTP API. The documented default Ollama endpoint is `http://localhost:11434`, and Ollama must be installed and running before the MCP client can use the server.

Ollama MCP Server MCP server is normally managed by the MCP client rather than started manually. After the client loads its configuration, it starts the Node.js entry point as a background subprocess. A client restart is required after changing the MCP configuration.

## Setup and configuration

The prerequisites are Node.js 18 or newer, npm 8 or newer, a running Ollama installation, and at least one locally downloaded Ollama model. The package can be installed globally or as a project dependency:

```bash
npm install -g @muhammadmehdi/ollama-mcp-server
```

The MCP configuration should run Node against the package's `dist/index.js` file. The repository examples set `OLLAMA_BASE_URL` to the local Ollama endpoint:

```json
{
  "mcpServers": {
    "ollama": {
      "command": "node",
      "args": ["<path-to-package>/dist/index.js"],
      "env": {
        "OLLAMA_BASE_URL": "http://localhost:11434"
      }
    }
  }
}
```

Global and local npm installations place the entry point in different directories, so the path in `args` must match the installation. Verify Ollama separately with its local API or the `ollama list` command before testing the client connection.

## Tools and capabilities

The supplied tool listing contains four model metadata capabilities:

- `name`: identifies a model.
- `size`: reports the model's size.
- `modified_at`: reports when the model was modified.
- `digest`: exposes the model digest.

These tools suit inventory and inspection tasks, such as checking which models are available locally and comparing their metadata. The provided tool listing does not establish additional operations for generating text, chatting, downloading models, or deleting models.

## Limitations and notes

Ollama MCP Server MCP server depends on a locally reachable Ollama process and does not replace Ollama itself. It also requires a local Node.js runtime and an MCP client that can launch stdio servers. The default endpoint is local, but the configuration supports specifying a different base URL through `OLLAMA_BASE_URL` when needed.

The README lists Windows, macOS, and Linux as supported platforms. Platform-specific setup still requires the appropriate shell, Node.js installation, and a correctly resolved path to the package entry point. The repository states that the project is released under the MIT license.

_Full upstream README: https://allmcps.com/mcp/muhammadmehdi-ollama-mcp-server/readme_