Skip to content

mcp.json, local models

Use Forkmate in LM Studio

LM Studio adds remote MCP servers, with headers, in mcp.json. Your model runs locally; Forkmate’s tools are called over the internet with your key.

Needs a key. LM Studio runs on your computer, so it connects with a personal key instead of a sign-in. Make a key in Forkmate · How keys work

No Forkmate account yet? Create one free first. It takes under a minute. You make your key there.

Example: in LM Studio, with Forkmate connected

I had two eggs and toast for breakfast.

Logged to Forkmate229 kcal
Egg, whole, cooked, hard-boiled · 2 large eggs (100 g)Bread, white · 1 slice (28 g)

Logged. Want me to total today so far?

Set up Forkmate in LM Studio

Forkmate’s server URL
https://mcp.forkmate.ai/
Streamable HTTP, with your personal key in the Authorization header.
  1. Make a keyIn Forkmate, open Settings → Keys for local apps (https://app.forkmate.ai/#keys), name the key after LM Studio, pick when it expires and click Make key. Copy it: Forkmate shows it once. How keys work: forkmate.ai/connect/keys/.
  2. Add the serverOpen the Program tab in the right sidebar → Install → Edit mcp.json and add the forkmate entry below, with your key in place of fmk_your_key_here.
  3. Pick a model that can call toolsLoad a model with tool-calling support, then enable Forkmate in the chat.
  4. Give your OK onceThe first time you log, Forkmate asks once for your OK to store your food diary. Agree the way it asks, in the chat or through the link it gives you to agree on the web, then log the meal again if it wasn’t saved.
  5. Log your first mealIn LM Studio, say “I had two eggs and toast for breakfast.” It logs the meal with calories and macros to your Forkmate diary.

mcp.json

{
  "mcpServers": {
    "forkmate": {
      "url": "https://mcp.forkmate.ai/",
      "headers": {
        "Authorization": "Bearer fmk_your_key_here"
      }
    }
  }
}

What you need

  • A free Forkmate account (sign up). There is no paid plan.
  • LM Studio 0.3.17 or later (MCP support).
  • The model runs on your computer, but Forkmate doesn’t: you need an internet connection, and what you log goes to your Forkmate diary.
  • Small local models call tools less reliably. If a meal doesn’t log, try a larger tool-capable model.

Checked against the vendor’s documentation: LM Studio docs: MCP. LM Studio changes its menus often; if yours looks different, trust what it shows.

What to say first

Talk to LM Studio the way you’d tell a friend:

  • “I had two eggs and toast for breakfast.”
  • “How many calories have I had today?”
  • “Make the toast two slices.”
  • “What did I eat this week?”

It logs the meal with calories, protein, carbs and fat to your diary at app.forkmate.ai.

Good to know in LM Studio

  • LM Studio shows a confirmation for each tool call, where you can check and edit the arguments first.
  • LM Studio’s mcp.json follows Cursor’s format, so a Cursor config pastes straight in.

If something goes wrong in LM Studio

Frequent context overflows with a local model
MCP tools add tokens, which LM Studio warns can overload a small model. Load a model with a longer context.
mcp.json breaks after pasting a config
Paste only the forkmate entry, not the outer mcpServers wrapper.

Checked against LM Studio 0.3.17 release notes, LM Studio docs: MCP.

Start logging from LM Studio

Create your free account, connect once, and your next meal is one sentence away. No card, no paid plan.

All assistantsStart free