Skip to main content
LM Studio runs local models and can also act as an MCP host. It supports remote MCP servers through an mcp.json file, edited from the Program tab and Perfai MCP is a remote server.
LM Studio’s own model is local and free to run — but tool-calling against an 18-tool MCP server needs a model that’s actually capable of structured tool use. Small quantized models can fail to respond at all.

Generate your Bearer token

Run this in your own terminal — substitute your Perfai email and password:
Copy the resulting string — you’ll paste it into mcp.json as Bearer <base64>.

Installation

  1. Open LM Studio and go to DeveloperLocal Server.
  2. Click the Program tab.
  3. This opens (or lets you edit) mcp.json. Add the following:
  1. Save the file. LM Studio reconnects to configured MCP servers automatically.

Verify the server is connected

  1. In the Program tab, check that perfai-cloud shows as connected with 18 tools listed.
  2. If it fails to connect, re-check the Authorization header value and confirm there’s no trailing whitespace in the base64 string.
A successful connection here only means LM Studio can see the server’s tools — it doesn’t guarantee your local model can actually call them.

Login

The prompts below run inside LM Studio’s chat, not your terminal. The only file-based step is editing mcp.json.
In LM Studio’s chat, type:
This calls login with no arguments, which opens a browser to Perfai’s own login page. Confirm the session with:

Register an app and run a scan


Fix and verify