Skip to content

docker-setup.md

Latest commit

 

History

History
85 lines (60 loc) · 2 KB

docker-setup.md

File metadata and controls

85 lines (60 loc) · 2 KB

Docker Setup

NEXO already ships a Dockerfile. This guide adds the missing persistent docker-compose.yml flow so you can keep the brain in a volume and expose a remote MCP endpoint when useful.

What the compose setup does

  • runs the NEXO server in a container
  • stores NEXO_HOME in a named volume
  • exposes http://localhost:8000/mcp
  • adds a container health check
  • still supports Claude Code and Codex through stdio via docker compose exec -T

1. Start the containerized runtime

docker compose up -d
docker compose ps

Defaults:

  • container name: nexo
  • host port: 8000
  • MCP path: /mcp
  • persistent volume: nexo_data

2. Verify the health check

docker compose ps
docker compose logs --tail=50 nexo

The service is healthy when the container is Up and the health column shows healthy.

3. Use the remote MCP endpoint from editor clients

Clients that support remote HTTP MCP can point at:

http://127.0.0.1:8000/mcp

This is the easiest path for Windsurf, Cursor, and other MCP IDEs that accept HTTP or Streamable HTTP servers.

4. Use the same container from Claude Code or Codex

Claude Code and Codex work best with stdio. Keep the container running and configure their MCP entry as:

{
  "command": "docker",
  "args": [
    "compose",
    "exec",
    "-T",
    "nexo",
    "python",
    "src/server.py"
  ]
}

That launches the same NEXO runtime inside the running container but keeps the client on the transport it expects.

5. Useful overrides

NEXO_MCP_PORT=8123 docker compose up -d

Environment knobs supported by the compose file:

  • NEXO_MCP_PORT
  • NEXO_MCP_PATH
  • NEXO_MCP_HOST
  • NEXO_MCP_TRANSPORT

6. When not to use Docker

If you want the full managed install, host-level client sync, launchers, and nexo chat, use npx nexo-brain directly on the host. Docker is best when you want:

  • a persistent local service endpoint
  • isolated runtime dependencies
  • one shared brain behind several editor clients