Sharing vscode continue extension and docker compose configs for tinkerers

[bradgillap]

Hello, this works well.

Just wanted to share. I got this working under docker compose and then was able to connect to it in vscode using the continue extension. Here's what the two important configurations look like. Hopefully it helps speed things up for someone else.

Docker Compose

services:
  mcp-outline:
    # Use the pre-built image or your locally built image
    image: ghcr.io/vortiago/mcp-outline:latest # Example image path
    ports:
      # Maps the container port 3001 to the host port 3001
      - "3001:3001"
    environment:
      # Enables HTTP transport for connectivity
      - MCP_TRANSPORT=sse
      # Replace with your actual Outline API Key
      - OUTLINE_API_KEY=your_api_key
      # Replace with the API URL of your self-hosted Outline instance
      - OUTLINE_API_URL=https://your-outline-instance.com/api

Continue MCP configuration

name: Outline Knowledge
version: 0.0.1
schema: v1
mcpServers:
  - name: Outline Search
    # CRITICAL CHANGE: Use the 'sse' type for remote HTTP connections
    type: sse 
    
    # Specify the URL of the running Docker container's SSE endpoint
    # This assumes you mapped container port 3001 to host port 3001
    url: http://your-dockercompose-server:3001/sse
    
    # Credentials are NOT needed here, as they are already injected
    # as environment variables into the Docker container.

I found that qwen3-coder:30b was actually sort of the best local LLM that could manage this as an agentic tool from the continue extension. Well, best given my resources anyway. I'm sure larger models may fair better as well but I had some interesting failures with gpt20b where it was less reliable. That new IBM Granite also fell flat for agentic tasks.

Anyway, really cool, thanks!

[Vortiago]

Cool, thanks for sharing. I haven't tried Continue myself.

I recently pushed a new version with a breaking change, where SSE is replaced with the newer Streamable HTTP as default and the port changing to 3000 since I noticed that was more commonly used. I tried to improve the readme as well taking inspiration from what you've written. :)