LM Studio Setup

LM Studio can run local models on your machine. With Automation Agent for Adobe Premiere, its most important role is usually Workflow AI: a saved workflow can call the local model for a bounded thinking step while the workflow is running.

This page is for the Local HTTP MCP transport in the Automation Agent panel.

Use this path when a local model is required or useful for your workflow. That may be because of company policy, sensitive transcripts or project data, or because you want to evaluate local-model behavior before using a remote agent.

If the difference between Assistant and Workflow AI is new to you, read AI Setup Roles first.

Which Role Does This Set Up?

Role	Supported with LM Studio?	Recommendation
Assistant	possible, but advanced / limited	Use Codex or Claude for most workflow authoring.
Workflow AI	yes	Recommended local/offline role for saved workflows.

The usual pattern is:

1. use Codex or Claude as the Assistant to build and test the workflow
2. design the saved workflow so only a small, clear AI task is handed to Workflow AI
3. configure LM Studio as the local Workflow AI provider
4. run the saved workflow later with the local model

For that pattern, see Developing Offline Workflows Online.

What This Setup Gives You

After Workflow AI setup, Automation Agent can call LM Studio's local server from a saved workflow.

This is useful for steps such as:

• proofreading transcript chunks
• summarizing selected text
• classifying markers, notes, or transcript sections
• extracting simple structured fields from text

The saved workflow should still handle the Premiere work: finding clips, reading transcripts, splitting long text, validating output, and writing results.

LM Studio can also be used as an MCP-capable chat client. In that mode, you can enable Automation Agent tools in an LM Studio chat and ask the local model to inspect or operate on the current Premiere project through the panel. That is an advanced path and usually needs more explicit instructions than Codex or Claude.

The Assistant-style LM Studio setup has two separate parts:

1. add Automation Agent to LM Studio's MCP server list
2. add the Automation Agent operating prompt through an LM Studio preset

The MCP server lets LM Studio call tools. The preset gives the model the basic operating rules for using Automation Agent safely.

The Workflow AI setup has a different focus:

1. start LM Studio's local server from the Developer tab
2. load a model
3. configure LM Studio as a Workflow AI provider in AI Setup
4. use Test & Save

What This Setup Does Not Give You

LM Studio does not currently get the same Automation Agent knowledge layer that Codex and Claude workflows can use.

In practice, that means the local model can see MCP tools and their input schemas, but it may not know enough about Automation Agent workflows by itself. It may not know which discovery steps to run first, how to write valid Runtime DSL programs, which Premiere data must be probed before use, or which examples are the right starting point for a specific task.

For simple tasks, the preset may be enough. For complex workflows, give LM Studio a detailed implementation prompt that includes the exact steps and code or DSL patterns it should adapt. A good way to create that prompt is to develop the workflow first with Codex, Claude Code, or another strong agent on synthetic or non-sensitive project material, then ask that agent to turn the working approach into an LM Studio implementation prompt.

For the broader pattern, see Developing Offline Workflows Online.

What You Need

Before configuring LM Studio as Workflow AI:

1. open Adobe Premiere Pro
2. open the Automation Agent panel
3. open AI Setup
4. open LM Studio
5. load a local model
6. start LM Studio's local server from the Developer tab

Use a local model that can handle the specific task your saved workflow will send it. Smaller models may be fine for short proofreading or classification tasks, but they may fail at long context or strict structured output unless the workflow keeps the handoff small.

If you also want to use LM Studio as an Assistant chat client, use a local model with strong tool-use behavior and follow the MCP/preset steps below.

Configure LM Studio As Workflow AI

In the Automation Agent panel:

1. open AI Setup
2. click Start Guided Setup
3. choose Local / offline
4. follow the LM Studio Workflow AI step

Or use the manual setup:

1. open AI Setup
2. open Workflow AI
3. choose or enter the LM Studio server URL, usually http://127.0.0.1:1234/v1
4. check the server and select a loaded model
5. use Test & Save

If the test fails, check that LM Studio is open, a model is loaded, and the local server is running in the Developer tab.

Optional: Install The MCP Config For LM Studio Chat

In the Automation Agent panel:

1. open MCP Server
2. open Agent Setup
3. select LM Studio
4. use Set Up or Update in the LM Studio mcp.json step

The panel writes the Automation Agent MCP entry to LM Studio's mcp.json file:

macOS / Linux: ~/.lmstudio/mcp.json
Windows:       %USERPROFILE%\.lmstudio\mcp.json

If you configured an Automation Agent auth token in MCP Server Setup, the panel also writes the matching Authorization header into the LM Studio MCP config. If you later change the token, update the LM Studio config from the panel again.

Optional: Install The Automation Agent Preset For LM Studio Chat

In the Automation Agent panel:

1. open MCP Server
2. open Agent Setup
3. select LM Studio
4. use Install or Update in the LM Studio preset step

The preset is written to:

macOS / Linux: ~/.lmstudio/config-presets/automation-agent.preset.json
Windows:       %USERPROFILE%\.lmstudio\config-presets\automation-agent.preset.json

When you start a new LM Studio chat, select the Automation Agent preset. Do not also paste the copied operating prompt into the same chat. The preset already contains that system prompt.

Start A Chat With Automation Agent Tools

If you want to use LM Studio as an MCP chat client, in LM Studio:

1. start a new chat
2. choose a local model with good tool-use support
3. select the Automation Agent preset
4. open the chat's tool options
5. enable the Automation Agent MCP server
6. ask the model to inspect before it changes anything

For example, start with a small read-only request:

Use Automation Agent to inspect the active Premiere project. First list what MCP tools you can use, then run a read-only probe and summarize the active sequence state. Do not modify the project.

If the model cannot list or call Automation Agent tools, check that the local MCP server is running, the mcp.json setup step completed successfully, and the Automation Agent MCP server is enabled in the chat's tool options.

If You Use LM Studio Through Its API Server

The normal setup above is for using LM Studio locally through its own chat interface.

LM Studio also has an API server that can be used by other local or network clients. That is a different security model. If you use Automation Agent tools through the LM Studio API server, LM Studio may require the server setting for allowing API clients to call MCP servers from mcp.json.

That LM Studio API setting is separate from the Automation Agent auth token:

• the LM Studio setting controls whether LM Studio API clients may use MCP servers listed in LM Studio's mcp.json
• the Automation Agent auth token controls whether Automation Agent accepts calls to its local MCP server

For the local LM Studio chat UI, you normally enable Automation Agent in the chat's tool options instead.

How To Work With Complex Recipes

For open-ended Premiere tasks, do not rely on a short prompt alone.

Prefer one of these patterns:

• use an Automation Agent workflow recipe that includes LM Studio implementation details
• provide a prompt that contains the exact MCP calls, Runtime DSL probes, validation steps, and approval points for the task
• develop the workflow with Codex or Claude Code on synthetic or non-sensitive material, then convert the proven workflow into a detailed LM Studio implementation prompt

A good LM Studio implementation prompt should include:

• what state to inspect first
• which Automation Agent MCP tools to call
• which Runtime DSL programs to validate before execution
• which values the model may adapt
• what output records to expect from probes
• when to stop and ask for approval
• how to verify the result after execution

This is more explicit than the prompts used for Codex or Claude Code because those clients can use richer Automation Agent guidance and examples.

Related Pages

• AI Setup Roles
• Agent Workflow via MCP
• Developing Offline Workflows Online
• Supported Agents And MCP Clients
• Ways To Run Automations
• High-Security and Restricted-Network Environments
• LM Studio MCP documentation