[Feature]: Support local Docker images

[marcindulak]

What feature would you like?

I would like to be able to use locally built Docker images in container-use.

I've tried this (and some other variations):

docker image pull nginx

Using default tag: latest
latest: Pulling from library/nginx
Digest: sha256:d5f28ef21aabddd098f3dbc21fe5b7a7d7a184720bc07da0b6c9b9820e97f25e
Status: Image is up to date for nginx:latest
docker.io/library/nginx:latest

docker tag nginx nginx-local

cu --version
container-use version 0.4.2 (89ebe77)

container-use config base-image set nginx-local:latest

container-use config show

Base Image:             nginx-local:latest
Workdir:                /workdir
Setup Commands:         (none)
Install Commands:       (none)
Environment Variables:  (none)
Secrets:                (none)

claude
> Read README.md

Output:

  ⎿  Error: failed to create environment: failed to resolve image 
     "docker.io/library/nginx-local:latest" (platform: "linux/amd64"): failed to resolve source 
     metadata for docker.io/library/nginx-local:latest: pull access denied, repository does not exist 
     or may require authorization: server message: insufficient_scope: authorization failed

Why would this be useful?

To avoid pushing (and pulling) large images to a cloud registry, or running a local Docker registry on the development machine.

[marcindulak]

It may be that running a local registry is not trivial. The unsuccessful example below is from Ubuntu 24 docker host, but I would like to also support at least macOS.

1. It looks like container-use (dagger?) enforces TLS encrypted traffic to a local registry, i.e. claude prints an error like this

   ⎿  Error: failed to create environment: failed to resolve image 
    "172.24.0.2:5000/nginx-local:latest" (platform: "linux/amd64"): failed to resolve 
    source metadata for 172.24.0.2:5000/nginx-local:latest: failed to do request: Head 
    "https://172.24.0.2:5000/v2/nginx-local/manifests/latest": dial tcp 172.24.0.2:5000: 
    i/o timeout
   

   This means that having an unencrypted local registry may not work with container-use.

2. After switching to a TLS for the local registry, self-signed root certificate used to sign TLS of the local registry may require starting a custom dagger-engine as mentioned here Corporate TLS issues #111 (comment). I've not been able to make this work yet.

   The registry with TLS traffic is started using this compose.yaml file

   services:
     registry:
       container_name: registry
       image: registry:2
       ports:
         - "127.0.0.1:5000:5000"
   
     caddy:
       container_name: caddy
       image: caddy:2
       ports:
         - "127.0.0.1:443:443"
       depends_on:
         - registry
       entrypoint: sh
       command:
         - -c
         - |
           echo '172.24.0.3 {
             reverse_proxy registry:5000
             tls internal
           }' > /etc/caddy/Caddyfile
           caddy run --config /etc/caddy/Caddyfile --adapter caddyfile
   

   docker compose up -d
   mkdir -p ca-certificates
   docker compose cp caddy:/data/caddy/pki/authorities/local/root.crt ca-certificates
   

   docker tag nginx:latest 127.0.0.1:5000/nginx-local:latest
   docker push 127.0.0.1:5000/nginx-local:latest
   

   When I start the custom dagger-engine manually

   docker run -d --rm \
    -v ./ca-certificates:/usr/local/share/ca-certificates \
    --name dagger-engine-custom \
    --privileged \
    registry.dagger.io/engine:v0.18.14
   

   to make sure the root.crt is present

   docker exec -it dagger-engine-custom sh -c "ls /usr/local/share/ca-certificates"
   84736cea.0  root.crt
   

   as soon as I start Claude Code, the custom dagger-engine gets replaced by

   docker ps | grep dagger
   bcdcacc0415f   registry.dagger.io/engine:v0.18.14   "dagger-entrypoint.s…"   2 minutes ago   Up 2 minutes   dagger-engine-v0.18.14
   

3. If a custom configuration of the container-use mcp server is used

   claude mcp remove container-use
   claude mcp add container-use -- bash -c "_EXPERIMENTAL_DAGGER_RUNNER_HOST=docker-container://dagger-engine-custom container-use stdio"
   

   REGISTRY_IP=$(docker inspect -f '{{range.NetworkSettings.Networks}}{{.IPAddress}}{{end}}' caddy)
   container-use config base-image set ${REGISTRY_IP}:443/nginx-local:latest
   # Output
   Base image set to: 172.24.0.3:443/nginx-local:latest
   

   with dagger-engine-custom having connectivity to the registry IP

   docker exec -it dagger-engine-custom sh -c "ping -c 1 ${REGISTRY_IP}"
   # Output
   PING 172.24.0.3 (172.24.0.3): 56 data bytes
   64 bytes from 172.24.0.3: seq=0 ttl=63 time=0.125 ms
   
   --- 172.24.0.3 ping statistics ---
   1 packets transmitted, 1 packets received, 0% packet loss
   round-trip min/avg/max = 0.125/0.125/0.125 ms
   

   Claude Code that uses the dagger-engine-custom, still fails with a dial tcp 172.24.0.3:443: i/o timeout error as above, despite the docker host being able to see the image in the local registry

   curl --cacert ca-certificates/root.crt -s https://172.24.0.3:443/v2/nginx-local/manifests/latest | head -1
   {
   ...