Control Mark Flow with an AI agent (MCP)

What the AI Agent (MCP) is

MCP (Model Context Protocol) is a common “socket” that lets AI apps such as Claude Desktop or Codex use outside tools. Once Mark Flow is connected through MCP, the AI becomes an “agent” (an assistant that does the work): you just ask, and the AI sends the commands to Mark Flow.

For example, you tell Claude: “Make 4 images of an astronaut cat, 16:9, with Google Flow”. Claude sends the command to Mark Flow; Mark Flow runs it in your own Chrome, with your own Google Flow account, and returns the images to Claude.

Before you start

• A Pro or Agency plan, signed in to Mark Flow.
• An AI app that supports MCP. This guide covers Claude Desktop and Codex.
• For Claude Desktop: Node.js LTS from nodejs.org. The configuration uses the npx command, which comes with Node.js.
• On Mark Flow 3.70.7 and older: turn on the Flow API (needs Mark Bridge) or Run queue in batches in Settings › General › Advanced, otherwise Google Flow generation commands fail. From the next version the AI Agent runs through the queue on its own, nothing to turn on. See Install Mark Bridge and turn on the Flow API.
• Mark Bridge is recommended: images and videos returned to the AI have their logo removed by Mark Bridge.

Create a token

A token is a secret string that tells Mark Flow the command comes from your AI.

1. Open the AI Agent (MCP) page: click the gear icon in the top-right corner of the side panel → Settings → choose AI Agent (MCP) in the left menu. Or, in the side panel, click the Tools tab → AI Agent.

   

2. Type a name you will recognise in the field to the left of Create token, for example “work computer”, then click Create token.

   

3. A green New token, shown only once box appears; the token starts with mfm_. Click Copy right away and paste it somewhere safe for now, for example Notepad.

4. The token appears in the list below with its name, never used or last used …, and its expiry date (one year later). The badge in the top-right corner of the page changes to the number of tokens you have.

Connect Claude Desktop

1. Install Node.js LTS if your computer does not have it. To check: open Command Prompt (Windows) or Terminal (Mac), type node --version and press Enter; if you see a version number, you are set.

2. Open Claude Desktop → Settings → Developer → click Edit Config. Claude opens claude_desktop_config.json or the folder that contains it.

3. Open that file with Notepad (Windows) or TextEdit (Mac). File location: Windows %APPDATA%\Claude\claude_desktop_config.json, Mac ~/Library/Application Support/Claude/claude_desktop_config.json.

4. If the file is empty or only contains {}, delete everything and paste the text below, then replace YOUR_TOKEN with your token (keep the word Bearer and one space before the token): {"mcpServers":{"markflow":{"command":"npx","args":["-y","mcp-remote","https://api.markflow.net/mcp","--header","Authorization:${AUTH_HEADER}"],"env":{"AUTH_HEADER":"Bearer YOUR_TOKEN"}}}}

5. Save the file. Quit Claude Desktop completely and open it again. Closing the window is not enough if Claude keeps running in the background: on Windows, find the Claude icon in the system tray next to the clock and choose to quit; on a Mac, press Cmd+Q.

6. Ask Claude: “list the markflow tools”. Claude answers with tools such as generate_image and generate_video. You can also see them under the + button in the bottom-left corner of the chat box › Connectors.

Connect Codex

1. Put the token in the MARKFLOW_MCP_TOKEN environment variable (a value your computer keeps for programs to read). Mac, Linux: open Terminal, run export MARKFLOW_MCP_TOKEN='YOUR_TOKEN' and start Codex from that same window; add the line to your ~/.zshrc file so you do not have to type it again. Windows: open PowerShell, run setx MARKFLOW_MCP_TOKEN "YOUR_TOKEN", then close and reopen the terminal window.

2. Open the file ~/.codex/config.toml (Windows: %USERPROFILE%\.codex\config.toml) in a text editor. If it does not exist, create a new file with exactly that name.

3. Add the four lines listed right after these steps at the end of the file, one per line. The easiest way: on the AI Agent (MCP) page, under Connect › Codex, click the Copy button below the configuration block and paste it into the file.

   

4. Save the file and restart Codex. Run codex mcp list or type /mcp in Codex: if you see markflow, it is connected.

• [mcp_servers.markflow]
• url = "https://api.markflow.net/mcp"
• bearer_token_env_var = "MARKFLOW_MCP_TOKEN"
• enabled = true

What the agent can do

• You do not need to remember tool names: just ask in plain words, for example “Check whether Mark Flow is ready” or “Make an 8-second video of ocean waves at sunset, 9:16”.
• The first time, Claude Desktop usually asks whether the markflow tools may run; allow them.
• Results: images show up right in the AI's chat, with download links valid for 48 hours. Mark Bridge has already removed the logo.
• Each account runs one command at a time; the next one waits until the previous one finishes.
• Reference images: give the AI a public image link starting with https:// that opens without signing in. PNG, JPG or WebP, up to 12 MB.
• Everything the agent makes shows up in the History tab with the label Generated via MCP (AI agent).

Save the agent's results to your computer

The Download when the agent generates card at the bottom of the AI Agent (MCP) page saves an extra copy of the results on your computer. It only applies to images and videos the agent makes with Google Flow.

1. Turn on the switch at the top of the card (off by default).

2. Folder: the sub-folder name, default markflow_mcp. Files go inside the main Mark Flow download folder, so Downloads/markflow_output/markflow_mcp if you kept the defaults.

3. Choose Image quality and Video quality (defaults 1K and 720p).

4. Click Save changes in the top-right corner.

Troubleshooting

Other Google Flow errors (out of generations, content refused…) are explained in Troubleshooting.