Tools, MCP & Pro Search

Tool calling, Pro Search, SearXNG, ephemeral MCP servers, LM Studio integrated MCPs, and code sandbox in LM Mini.

Answer first

Tools let the model call capabilities — web search, time, MCP servers, code sandbox — then continue with real results. Turn them on globally or per chat / group participant.

Master switches

Settings → Advanced → Tools:

Composer chips toggle Pro Search / tools for a turn without digging through Settings.

  1. Sign in with Pro where required
  2. Enable Pro Search in Tools
  3. Use a tool-capable model (browse UIs may show a Pro Search badge)
  4. Ask something that needs the live web

The app runs the tool loop, shows activity in the transcript, then continues the reply.

Tip

If search “does nothing,” confirm tool use is on, the model supports tools, and you are not on a text-only path. Prefer a tool-capable instruct model.

SearXNG

Point LM Mini at a SearXNG instance you control when you want search without Pro Search. Prefer HTTPS. Failures should show as tool errors — update the app if search fails silently.

Ephemeral MCP

Add a label + URL (+ Bearer / custom headers). LM Mini sends them on compatible LM Studio requests. Only add servers you trust.

Integrated LM Studio MCPs

Configure tools in LM Studio, then enable the matching names in LM Mini. These run on the LM Studio machine, not as a URL you paste in the app. Ollama does not host those LM Studio plugins.

Group chat & Personas

Enable tools per participant. Personas do not force tools on — the chat/group toggle still matters.

Safety