Hostly

Connect LM Studio ​

LM Studio runs language models locally on your computer and can use Hostly as a remote MCP server. Unlike ChatGPT and Claude, LM Studio needs the Hostly OAuth client credentials to sign you in. These credentials are public and only allow the sign-in with your own Hostly account.

INFO

This guide requires an LM Studio version with the MCP settings UI. LM Studio may adjust the labels below over time.

Add Hostly ​

  1. Open Settings in the bottom left corner of LM Studio.

  2. In the Integrations section, select MCP.

  3. Click Add MCP Server Manually.

  4. Fill in the connection details:

    • Name: Hostly

    • Connection: choose Web address

    • Server address:

      text
      https://api.hostly.ninja/mcp

    • Authentication: keep Automatic

  5. Under Advanced settings, turn on Show advanced options and enter the Hostly client credentials under OAuth client details:

    • Client ID:

      text
      tpc_b5fQnmELaqpqmyumDPdnHS

    • Client secret:

      text
      -x2E5oxx5jry-Jgqi__ULGdzviJGtNiD1AAzaZl5cuFkyPDXEEYz6HioM5GBU5z1

  6. Keep OAuth scopes on Automatic and leave the other advanced options unchanged.

  7. Save the server and sign in with your Hostly account in the browser window that opens.

Once signed in, the Installation section shows a Sign out button for the Hostly server.

Use Hostly in a chat ​

Load a model that supports tool use, enable Hostly for the chat, and ask, for example:

text
Show me the logs of the service "api" in my workspace.

LM Studio asks for your approval before it calls a Hostly tool. Review tool calls that change or delete something before you approve them.

Choose a model ​

Hostly provides a few dozen tools. Their descriptions take up part of the model's context window before the conversation even starts, which can overwhelm small local models. If the model ignores the tools, picks the wrong one, or cuts off its answers:

  • Turn off the Hostly tools you don't need for the current task, for example everything except listing services and reading logs.
  • Increase the context length when you load the model.
  • Use a larger model with good tool-use support.

Troubleshooting ​

  • Sign-in fails: Check that the client ID and client secret are copied completely and that Authentication is set to Automatic. Then click Sign out in the Installation section and connect again.
  • The model does not use Hostly: Make sure the model supports tool use and Hostly is enabled for the chat. Then see Choose a model.
  • A tool call is rejected: Your Hostly account lacks the required permission in that workspace. Ask a workspace member with the right permissions to adjust your role.
  • Start over: Click Remove MCP in the Installation section and add Hostly again.