cyberneticlibrary
← library

Add Ollama MCP server integration

add-ollama-toolskillsetup L2★29,740
nanocoai/nanoclaw ↗
What it does

Add Ollama MCP server exposing local models as agent tools

Best for

When you want the agent to offload work to local LLMs (inference, summaries) and optionally manage the model library.

Inputs
  • · Ollama host URL (default localhost:11434)
  • · OLLAMA_ADMIN_TOOLS flag (optional)
Outputs
  • · registered mcp__ollama__* tools: list_models, generate, pull, delete, show, list_running
Requires
  • · Ollama daemon
  • · stdio MCP protocol
  • · Bun runtime in agent-runner
Preconditions

Phase 1: Ollama running with models; Phase 2: code copied; Phase 3: wiring + rebuild + test

Failure modes
  • · Ollama daemon unreachable
  • · Model load timeout (30s cold start)
  • · Registration missing in index.ts
  • · env forwarding broken
Trust signals
  • · 7-phase implementation with guard tests (ollama-registration.test.ts, ollama-wiring.test.ts)
  • · Keyless, config-only setup
  • · Covers cold-start timing edge case