MCP Servers not loading for models with cloud-proxy backend

[popsUlfr]

Hello,

models using cloud-proxy backend and specifying mcp servers, have them not load and selectable in chat.

LocalAI version:
docker.io/localai/localai:master-gpu-hipblas

Environment, CPU architecture, OS, and Version:
uname -a : Linux fwd 7.1.3-2-cachyos #1 SMP PREEMPT_DYNAMIC Wed, 08 Jul 2026 06:14:22 +0000 x86_64 GNU/Linux
OS: CachyOS
Server: Framework Desktop AMD RYZEN AI MAX+ 395 w/ Radeon 8060S with 128GB ram

Describe the bug
Specifying mcp servers in a model using cloud-proxy backend does not make them selectable in chat.
The server widget is forever loading.

To Reproduce
+ Add Model > OpenAI Cloud Proxy > Fill out > Create Model > Add the MCP servers in yaml > go chat with the model > try to select the mcp servers > forever loading

Expected behavior
MCP servers should appear in MCP widget on chat and be selectable

Logs

Additional context

backend: cloud-proxy
known_usecases:
    - chat
name: qwen3.6
pii:
    enabled: true
proxy:
    api_key_env: OPENAI_API_KEY
    mode: passthrough
    provider: openai
    request_timeout_seconds: 120
    upstream_model: qwen3.6
    upstream_url: https://<omitted>/v1/chat/completions
mcp:
    remote: |
        {
          "mcpServers": {
            "search-engine": {
              "url": "http://host.containers.internal:31094/mcp"
            }
          }
        }
    stdio: |
        {
          "mcpServers": {
            "memory": {
              "command": "docker",
              "args": [
                "run", "-i", "--rm", "-v", "localai-mcps:/data",
                "ghcr.io/mudler/mcps/memory:master"
              ]
            }
          }
        }