› gflow-cli ★ Star on GitHub

Open source · Beta · MCP-native

The MCP server and CLI for Google Flow.
Drive it from your terminal, or let your AI assistant drive it.

gflow-cli is an open-source MCP server and command-line tool for Google Flow: generate Veo 3.1 and Omni video and Nano Banana images instead of clicking through the web UI. Point Claude, Cursor, or Copilot at it and the agent picks the models, prompt tooling, and settings for you.

$ uv tool install gflow-cli Get started → ▶ Watch the demo

Latest version on PyPI GitHub stars Last commit MIT licensed

The agent-native part

You describe it. The agent runs it.

gflow-cli ships an MCP server, so an AI assistant connects to it natively — no pasting docs, no copying commands back and forth. CLI-to-MCP parity is enforced in CI: anything the terminal can do, your assistant can do.

  1. 01
    gflow mcp setup
    One command writes the server entry into Claude Desktop, Cursor, or VS Code.
  2. 02
    Ask in plain language
    The agent picks the Veo, Omni or Nano Banana model, the aspect and the prompt tooling.
  3. 03
    Get files back
    It runs the generation on your own Flow session and hands you the output.

What it handles for you

Built for real runs, not just a demo.

The knobs an agent (or you) leans on, plus the reliability work that keeps batch jobs and big projects from falling over.

--tool creative-director

Prompt-crafting, built in

Rewrites a terse line with Google's 5-component formula before a single credit is spent.

gflow instructions

Instruction cards

Persistent, credit-free brief cards that steer look and tone across every generation.

reuse by UUID

Reference an asset, don't re-upload

Chain a still into a video by its id — no duplicate uploads, no shuffling files.

gflow character create

The same face, shot after shot

Mint a Flow Character once and the same person carries across every generation — instead of re-rolling until the face matches.

exit 23

A broken run costs a message, not credits

A billed request is sent once and never retried automatically, and a drifting UI aborts with a distinct exit code instead of resubmitting into a stale state. Nothing generated, nothing spent.

nightly drift probe

It notices when Flow moves

A selector registry plus a nightly CI probe against the live UI, so breakage is caught by a scheduled run rather than by your batch job.

GFLOW_CLI_JITTER_RANGE

Configurable batch pacing

Tunable jitter between submissions so multi-prompt runs stop tripping rate limits.

movie.toml

Multi-scene pipeline

Direct a whole sequence from one manifest — chain clips, compose scenes, render.

Read this before you install

Independent, in beta, and honest about it.

  • Not affiliated with Google. gflow-cli is an independent project. Flow, Veo, and Imagen are Google's.
  • Reverse-engineered. It drives Flow's private interfaces, so Google can change them without notice. Every release since v0.54.0 is live-verified against real Flow.
  • Runs a headed browser on your own Flow session. Treat it as your own account risk.
  • Needs a Google account with Flow access. Every video generation bills your own account's Flow credits; images spend none but count against a daily image quota.

FAQ

Common questions

Is gflow-cli an official Google tool?

No. It is an independent, MIT-licensed project that drives your own signed-in Google Flow session.

Do I need a Gemini or Vertex AI API key?

No. gflow-cli uses your Google account's Flow access and Flow credits, not the pay-per-use API.

How do I connect Claude or Cursor to Google Flow?

Install with uv tool install gflow-cli, sign in once with gflow auth login --browser chrome, then run gflow mcp setup (Claude Desktop), gflow mcp setup --target cursor, or install the Claude Code plugin.

Can an AI agent spend my Flow credits without asking?

Not when the server runs as gflow mcp run --no-spend: the credit-spending tools are never registered. Image generation spends no credits (it counts against a daily image quota); only video spends credits.