Skip to main content

Supported Agents And MCP Clients

Automation Agent for Adobe Premiere is designed to work with the agent or model environment that fits your workflow.

The MCP server is the connection layer. It lets external agents inspect the current Premiere project, validate Runtime DSL, and execute permission-scoped workflows through the Automation Agent panel.

The practical question is not only "which clients support MCP?" It is:

  • which agent is strong enough for the workflow?
  • which model or provider is allowed for the material?
  • does the workflow need live reasoning during execution?
  • should the result become a saved script or implementation prompt for reuse?

If you have not made that decision yet, start with Ways To Run Automations.

If you are not sure what Assistant and Workflow AI mean, start with AI Setup Roles. In short: the Assistant is the AI you talk to, while Workflow AI is a provider a saved workflow can call later.

This page is only about MCP clients. If you want to launch a saved script from a shortcut, Stream Deck, or Touch Portal, see Keyboard Shortcuts and Remote Triggers instead.

The important distinction is not just whether a client supports MCP in principle, but which transport path that client expects:

  • some clients work best with the panel-managed HTTP MCP server
  • some clients work best with a client-launched local stdio runtime

Currently Supported and Documented​

These are the client paths that are currently supported and documented as first-class workflows:

  1. OpenAI Codex
  2. Claude Code
  3. Claude Desktop Chat
  4. LM Studio

Role Overview​

Tool or surfaceAssistantWorkflow AINotes
Codex desktop appyesmacOS often includes the runnable Codex binary; Windows usually needs separate CLI setupUse the Codex skill and MCP config for Assistant work.
Codex CLIusually through the same Codex setupyesNeeds codex login before saved workflows can call it.
Claude Desktop Code tabyesnoThis is the recommended Claude Desktop surface for HTTP MCP Assistant work.
Claude Code CLIoptional / advancedyesNeeds claude auth login; used as the Claude Workflow AI provider.
Claude Desktop Chat tabyes, with limitationsnoUses the separate stdio desktop-extension path.
Claude Desktop Cowork tabnonoNot currently supported for local Automation Agent MCP workflows.
LM Studioadvanced / limitedyesBest used as local Workflow AI, especially for offline or approved-model workflows.

OpenAI Codex​

Recommended when:

  • you want a strong OpenAI-based coding agent for technical Automation Agent workflows
  • you have access to Codex, including limited access on some ChatGPT plans
  • you want the generated Automation Agent Codex skill and MCP setup to guide discovery, validation, and Runtime DSL authoring
  • you may also want Codex CLI as Workflow AI for saved workflows

Transport path:

  • panel-managed HTTP MCP server

Role setup:

  • Assistant: configure Codex MCP and install the Automation Agent Codex skill
  • Workflow AI: configure and test Codex CLI in AI Setup

See:

Claude Code​

Recommended when:

  • you want a strong Claude-based coding agent for technical Automation Agent workflows
  • you have access to Claude Code
  • you want a guided setup with the generated Automation Agent Claude Skill
  • you may also want Claude Code CLI as Workflow AI for saved workflows

Transport path:

  • panel-managed HTTP MCP server

Role setup:

  • Assistant: configure Claude Code MCP and install or confirm the Automation Agent Claude Skill where your Claude surface supports it
  • Workflow AI: configure and test Claude Code CLI in AI Setup

See:

Claude Desktop Chat​

Recommended when:

  • you want a more end-user-facing Claude Desktop chat workflow
  • you can accept that, unlike Codex or Claude Code, this client cannot access your project folder directly
  • you still want Claude to manipulate files through the Automation Agent MCP workflow

Transport path:

  • Claude Desktop-launched local stdio runtime

Role setup:

  • Assistant: install the MCPB desktop extension and upload the Claude Skill
  • Workflow AI: not supported through Claude Desktop Chat; use Claude Code CLI if you want Claude as Workflow AI

See:

LM Studio​

Recommended when:

  • you want to run the model locally
  • your material or organization requires a local or approved model path
  • you want a local MCP-capable chat client
  • you can provide more explicit workflow instructions for complex Premiere tasks
  • you want a local Workflow AI provider for saved workflows

Transport path:

  • panel-managed HTTP MCP server

Role setup:

  • Workflow AI: start the LM Studio local server, load a model, and configure LM Studio as a provider in AI Setup
  • Assistant: possible for advanced users, but Codex or Claude is usually better for creating complex Automation Agent workflows

See:

Important expectation:

  • LM Studio can use Automation Agent as an MCP tool server, but it does not currently receive the same bundled Automation Agent guidance and examples that Codex and Claude workflows can use through their guided setup. For complex workflows, use prepared workflow recipes or detailed implementation prompts.

This is why reusable implementation prompts matter. A local model may be allowed to process sensitive material, but it usually benefits from a more explicit, tested workflow plan.

Currently Not Supported as a First-Class Workflow​

Claude Desktop Cowork​

The Cowork tab is currently not a supported local Automation Agent workflow.

At the moment, the documented Claude Desktop local path is:

  • Chat tab + MCPB desktop extension + Claude skill ZIP

The documented technical Claude path is:

  • Code tab / Claude Code + HTTP MCP config

Other MCP-Capable Clients​

Other clients may still work, but they are currently manual / advanced setups, not first-class documented workflows.

Examples include:

  • Cursor
  • Windsurf
  • GitHub Copilot coding agent
  • other MCP-capable IDE or agent shells

For those clients, the key question is:

  • does the client expect an HTTP MCP server?
  • or does it expect a local stdio process?

If the client expects HTTP​

Use the panel in:

  • Local HTTP MCP transport

That means:

  • start the panel-managed MCP server
  • use the local HTTP MCP URL
  • configure the client manually with the matching MCP entry

If the client expects stdio​

Use the panel in:

  • stdio MCP transport

That means:

  • do not use the panel-managed HTTP server path
  • let the client launch the local stdio runtime itself
  • configure that client manually according to its own local MCP process model

Important note for manual clients​

For Cursor, Windsurf, Copilot, and similar tools, Automation Agent does not currently ship dedicated guided installers in the panel.

Those clients may still be usable, but the MCP setup is currently manual and depends on that client's own MCP support model.

Practical Recommendation​

If you want the most reliable current paths:

  1. Use OpenAI Codex for the main OpenAI technical workflow.
  2. Use Claude Code for the main technical Claude workflow.
  3. Use Claude Desktop Chat for the simpler end-user-facing Claude Desktop workflow.
  4. Use LM Studio when you need a local model and can work with more explicit prompts.

If you are evaluating another MCP-capable client, first determine whether it needs:

  • HTTP -> use Local HTTP MCP
  • stdio -> use stdio MCP

Then configure that client manually.

For repeated workflows, do not rely on a broad one-line prompt every time. Develop the workflow once with a strong Automation Agent-aware client, then turn the successful run into a reusable implementation prompt or saved script. See Create Your Own Recipes.