claude-sidekick

Claude Sidekick is an MCP server that connects Claude to your local Ollama models, enabling you to offload simpler tasks to local models and preserve Claude tokens for more complex work. Once set up, you can access a suite of Ollama-based tools directly from Claude, including text generation, chat, embeddings, code generation, and summarisation. The server exposes tools such as ollama_generate_text for straightforward writing tasks, ollama_chat for interactive Q&A with local models, ollama_embed_text for semantic embeddings, ollama_code_generation for boilerplate or small coding tasks, ollama_summarise for concise or extended summaries, and management helpers like ollama_list_models and ollama_pull_model. To use these tools, configure Claude Desktop with the MCP server and restart Claude to load the tools into your workspace. You can then issue natural language prompts requesting a specific Ollama model or task, e.g., generate a brief email, write a quick script, or create a document summary, and Claude will route the request to the appropriate local Ollama model through the MCP server.

Prerequisites include having Ollama installed and running on your machine, and having Node.js tooling available to run the MCP server. The available tools provide a flexible workflow: generate text for simple content, converse with local models via chat, compute embeddings for semantic search, and generate code when you need quick boilerplate. You can also list models or pull new ones into Ollama as your needs evolve, all through the same MCP-enabled interface within Claude.

Prerequisites

• Ollama installed and running locally
• Node.js and npm installed
• Basic familiarity with editing Claude Desktop MCP configuration

1. Install Ollama (if not already installed)

• macOS: brew install ollama
• Linux: curl -fsSL https://ollama.ai/install.sh | sh
• Windows: Download from https://ollama.ai/download and install

2. Install the MCP server dependencies

• Create a dedicated project folder and install dependencies

mkdir mcp-claude-sidekick
cd mcp-claude-sidekick
npm install

3. Prepare the server structure

• Move your MCP server entry into a src directory and build

mkdir src
# Place your index.ts (or index.js) into src/ as appropriate
# If you're starting from a template, configure the build to output to dist/

4. Build the project

npm run build

5. Configure Claude Desktop to load the MCP server

• Edit Claude Desktop MCP config (macOS): ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%/Claude/claude_desktop_config.json

Example config snippet (replace with the actual built path):

{
  "mcpServers": {
    "claude-sidekick": {
      "command": "node",
      "args": ["/absolute/path/to/your/claude-sidekick/dist/index.js"],
      "env": {}
    }
  }
}

6. Restart Claude Desktop and verify the MCP server appears and is usable.

Notes

• Ensure the path to the built JavaScript file is absolute and correct.
• If you modify the server, rebuild and restart Claude Desktop to reload changes.

Tips and common issues:

• Environment variables: You can override Ollama settings with OLLAMA_BASE_URL (default http://localhost:11434) and OLLAMA_TIMEOUT (ms).
• Ensure Ollama is running and accessible (e.g., curl http://localhost:11434/api/tags).
• If Claude does not surface the tools, verify the MCP config path, absolute path to dist/index.js, and that Claude Desktop was fully restarted.
• For performance, use smaller models for simple tasks (e.g., llama3.2:1b) and keep frequently used models warm with occasional Ollama runs.
• When pulling new models via Ollama, ensure the model names are correct and that you have sufficient disk space.
• Security: This setup runs locally; no external network calls are required, and no API keys are needed.