CodeBuddy integrates with Docker for running local AI models and managing MCP servers. This covers three capabilities: Docker Model Runner for GPU-accelerated inference, Ollama containers via docker-compose, and the MCP Docker Gateway for unified tool catalogs.

Docker Model Runner

Docker Model Runner is Docker Desktop’s built-in inference engine. CodeBuddy can enable it, pull models, and route requests to it automatically.

Setup

1. Ensure Docker Desktop is installed and running
2. In CodeBuddy, use the Docker panel or ask the agent: “Enable Docker Model Runner”

This runs:

docker desktop enable model-runner --tcp=12434

Models are then accessible at http://localhost:12434/engines/llama.cpp/v1 (OpenAI-compatible API).

Model management

Action	CLI equivalent
List models	docker model ls --json
Pull a model	docker model pull <name>
Delete a model	docker model rm <name>
Use a model	Auto-configures local.model, local.baseUrl, and switches to Local provider

Port fallback

When you select “Use Model”, CodeBuddy tries Docker Model Runner on port 12434 first (1-second timeout). If unavailable, it falls back to Ollama on port 11434. The appropriate local.baseUrl is set automatically.

Ollama via Docker Compose

CodeBuddy bundles a docker-compose.yml for running Ollama in a container:

# Key settings:
# - 32 GB memory limit
# - GPU passthrough (optional)
# - Persistent volume for models
# - Port 11434 exposed

Start with: “Start Ollama container” or via the Docker panel.

docker compose -f <extension-path>/docker-compose.yml up -d

Ollama model management

Action	CLI equivalent
Check status	docker ps --filter name=ollama --format json
List models	docker exec ollama ollama list
Pull a model	docker exec ollama ollama pull <name>

Docker availability detection

CodeBuddy probes the Docker socket directly (/var/run/docker.sock on macOS/Linux, \\.\pipe\docker_engine on Windows) to check if Docker is running. Results are cached for 15 seconds to avoid repeated spawn failures.

MCP Docker Gateway

The MCP Docker Gateway provides a unified catalog of MCP tools from a single Docker-managed endpoint.

How it works

1. CodeBuddy detects the Docker Gateway via docker mcp --help + docker info
2. When docker-gateway is the only enabled MCP server, all MCP tools come from one unified catalog
3. Communication uses stdio or SSE transport

Reliability features

Feature	Behavior
Circuit breaker	3 failures → 5-minute cooldown per server
Idle shutdown	Gateway disconnects after 5 minutes of inactivity, reconnects on-demand
macOS PATH fixup	Injects /usr/local/bin:/opt/homebrew/bin etc. for proper CLI discovery
Graceful degradation	If Docker is unavailable, the agent continues with core tools only

Docker panel commands

The Docker panel in the CodeBuddy webview exposes 10 operations:

Command	Description
docker-enable-runner	Enable Docker Model Runner
docker-start-compose	Start Ollama container
docker-check-status	Check Docker availability
docker-check-ollama-status	Check Ollama container status
docker-get-models	List all local models
docker-get-local-model	Get current local model config
docker-pull-model	Pull a Docker model
docker-pull-ollama-model	Pull an Ollama model
docker-delete-model	Delete a model
docker-use-model	Switch to a model (with port fallback)

Terminal security

Docker commands executed through the terminal are strictly whitelisted. Only these argument patterns pass validation:

• Model operations: model ls, model pull <name>, model rm <name>
• Container operations: ps --filter name=ollama, compose up -d, exec ollama ollama list
• MCP operations: mcp server ls, mcp gateway run
• Setup: desktop enable model-runner --tcp=12434

Model names are validated against /^[a-zA-Z0-9/:\-.]+$/ to prevent injection.

Configuration

Docker integration uses existing local model settings — there are no Docker-specific settings:

Setting	Description
local.model	Model name (auto-set when using “Use Model”)
local.baseUrl	Endpoint URL (auto-set with port fallback)
generativeAi.option	Set to "Local" when switching to a Docker model

See Local Models for the full local model setup guide.