ollama-agent-mcp

MCP stdio server that delegates bounded local analysis to an Ollama model.

This package is intentionally named after Ollama because it calls the Ollama HTTP API directly. The repository and build MCPs in this toolkit do not require local models.

Other local model runtimes exist, including llama.cpp servers, LM Studio or Jan with local OpenAI-compatible endpoints, LocalAI, vLLM, and MLX-based servers. Those would be better served by a separate provider package or a future generic local-model MCP. This package keeps the first version small and explicit.

Requirements

• Node.js 22+
• Ollama running locally or reachable from your machine
• A pulled local model, for example:

ollama pull qwen3.5:9b

Install

npm install -g @erivelto_muller/ollama-agent-mcp

Run

ollama-agent-mcp

The legacy binary alias agent-ollama is also included.

Configuration

Set LOCAL_AGENT_MCP_CONFIG to a YAML file:

security:
  allowed_roots:
    - /path/to/repository
repository:
  default_root: /path/to/repository
ollama:
  endpoint: http://127.0.0.1:11434
  model: qwen3.5:9b
  context_size: 8192
  temperature: 0.1

Environment overrides:

• OLLAMA_ENDPOINT
• OLLAMA_MODEL
• LOCAL_AGENT_ALLOWED_ROOTS

Tools

• ollama_health
• summarize_source_files
• classify_search_matches
• analyze_git_diff_locally
• compare_implementations