add-ollama-tool

name	add-ollama-tool
description	Add a local Ollama MCP server so the Codex worker can call local models for cheaper/faster summarization, translation, or general helper tasks.

Add Ollama Integration

This skill adds a stdio MCP server that exposes local Ollama models to the local Codex worker.

Tools added:

• ollama_list_models — lists installed Ollama models
• ollama_generate — sends a prompt to a model and returns the response

Current Runtime Note

NanoClaw now runs a local Codex worker, not a Docker container. Avoid Docker-only assumptions such as host.docker.internal, per-group agent-runner-src copies, or rebuilding container/build.sh.

The skill package may still contain old container-era modify/ files. If an apply operation conflicts, port the intended changes into the current local-worker files:

• container/agent-runner/src/ollama-mcp-stdio.ts
• container/agent-runner/src/runtime/codex-runtime.ts
• src/container-runner.ts only if host log surfacing is still needed

Phase 1: Pre-flight

Check if already applied

Read .nanoclaw/state.yaml. If ollama is in applied_skills, skip to configuration and verification.

Check prerequisites

Verify Ollama is installed and running on the host:

ollama list

If Ollama is not installed, direct the user to https://ollama.com/download.

If no models are installed, suggest pulling one:

ollama pull gemma3:1b
ollama pull llama3.2
ollama pull qwen3-coder:30b

Phase 2: Apply Code Changes

Initialize the skills system if needed:

npx tsx scripts/apply-skill.ts --init

Apply the skill:

npx tsx scripts/apply-skill.ts .claude/skills/add-ollama-tool

Expected outcome:

• add container/agent-runner/src/ollama-mcp-stdio.ts
• register an ollama stdio MCP server in the Codex runtime's mcp_servers config
• allow the worker to reach the host Ollama endpoint through OLLAMA_HOST
• record the application in .nanoclaw/state.yaml

Validate:

npm run build

Phase 3: Configure

By default, local worker processes can use:

OLLAMA_HOST=http://127.0.0.1:11434

Add a different OLLAMA_HOST to .env only when Ollama is not listening on localhost.

Restart the service:

launchctl kickstart -k gui/$(id -u)/com.nanoclaw
# Linux: systemctl --user restart nanoclaw

Phase 4: Verify

Tell the user to ask from a registered chat:

use ollama to list available local models

Then monitor:

tail -f logs/nanoclaw.log | grep -i ollama
ls -lt groups/*/logs/worker-*.log | head

Troubleshooting

Failed to connect to Ollama

1. Verify Ollama is running: ollama list
2. Check OLLAMA_HOST in .env
3. From the host, test: curl -s http://127.0.0.1:11434/api/tags

Agent does not see Ollama tools

1. Check that container/agent-runner/src/runtime/codex-runtime.ts includes the ollama MCP server entry.
2. Rebuild with npm run build.
3. Restart the service.
4. Inspect the newest groups/<group>/logs/worker-*.log for MCP startup errors.