Add MCP Ollama

Add Ollama MCP Plugin

Step 0: Language selection

Before proceeding with any other steps in this skill, ask the user which language to continue in using AskUserQuestion. Keep this initial prompt in English because the preferred language is not yet known.

• Question: Which language should I continue in?
• Options: English, 日本語 (Japanese)

Use the selected language for all subsequent user-facing messages and for every further AskUserQuestion prompt in this skill. Do not translate code, file paths, shell commands, or file contents.

This skill configures the Ollama MCP plugin so container agents can use local LLM models (llama3.2, mistral, gemma2, etc.) for cheaper/faster tasks.

UX Note: Use AskUserQuestion for all user-facing questions.

Phase 1: Pre-flight

Check Ollama is installed

which ollama 2>/dev/null && ollama --version || echo "NOT_FOUND"

If not installed, AskUserQuestion: Install Ollama?

• macOS: brew install ollama
• Linux: curl -fsSL https://ollama.com/install.sh | sh
• Or download from https://ollama.com

Check Ollama is running

curl -s http://localhost:11434/api/tags | head -1 || echo "NOT_RUNNING"

If not running: ollama serve & or start the Ollama app.

Check models are installed

ollama list

If no models, suggest pulling one:

ollama pull llama3.2

Phase 2: Configure entry.ts

Verify deploy/{ASSISTANT_NAME}/host/entry.ts contains the Ollama MCP plugin registration. If not, add this block after the orchestrator creation:

orchestrator.registerMcpPlugin("ollama", {
  entryPoint: "/app/mcp-plugins/ollama/dist/index.js",
});

No API token needed — Ollama runs locally. The container reaches the host via host.docker.internal:11434.

If deploy/{ASSISTANT_NAME}/host/entry.ts is outdated, compare with deploy/templates/host/entry.template.ts and update accordingly.

Custom Ollama host (optional)

If Ollama runs on a different host/port, pass it as an environment variable:

orchestrator.registerMcpPlugin("ollama", {
  entryPoint: "/app/mcp-plugins/ollama/dist/index.js",
  env: { OLLAMA_HOST: "http://192.168.1.100:11434" },
});

Phase 3: Rebuild & Verify

Rebuild Docker image (if not already built with Ollama plugin)

Check which agent type is configured (CONTAINER_IMAGE in .env), then rebuild the matching image:

Claude Code:

./container/claude-code/build.sh

Open Code:

./container/open-code/build.sh

Restart nagi

pnpm dev

Test

Tell user:

Send a message in your Slack channel:

• "What models does Ollama have?"
• "Use Ollama with llama3.2 to summarize: The quick brown fox jumps over the lazy dog."

The agent should use mcp__ollama__ollama_list_models or mcp__ollama__ollama_generate.

Available Tools

Once configured, container agents have access to:

• ollama_list_models — List installed local models with sizes
• ollama_generate — Send a prompt to a local model and get a response

Troubleshooting

"Failed to connect to Ollama"

1. Ollama must be running on the host: ollama serve or start the Ollama app
2. Verify: curl http://localhost:11434/api/tags
3. Docker must be able to reach the host — host.docker.internal is used automatically on macOS/Windows. On Linux, ensure --add-host=host.docker.internal:host-gateway is set (nagi handles this automatically).

Agent doesn't see Ollama tools

1. Check deploy/{ASSISTANT_NAME}/host/entry.ts has registerMcpPlugin("ollama", ...)
2. Check Docker image was rebuilt: ./container/claude-code/build.sh (or ./container/open-code/build.sh for Open Code)
3. Restart nagi

No models available

Pull a model on the host:

ollama pull llama3.2    # 2GB, fast
ollama pull mistral     # 4GB, good quality
ollama pull gemma2      # 5GB, Google's model