Provides tools for Google Flow, enabling AI agents to generate videos with Google Veo and images with Google Imagen, plan multi-scene batches, check credits, and stitch generated clips into a final video.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@io.github.ethanhamilthon/yenflow Plan a 3-scene video batch and estimate credits"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Stateless Model Context Protocol (MCP) server and automation engine for Google Flow (Veo video and Imagen generation).
Designed for AI agents (Claude, Cursor, Pi, Zed, Windsurf) to orchestrate professional video and image generation with zero local database requirements and zero window disruptions.
Stateless MCP Core: The MCP client maintains the entire workflow state machine (prompts, scene sequence, hashes, approvals, and media IDs). The server requires no local database, migrations, or persistent disk lock-in.
Silent Background Execution: Driven entirely through headless Playwright (headless=True, background=True). Zero visible Chrome windows, zero user focus stealing on macOS/Linux.
Two-Phase Credit Lock: Paid video generation requires generating a deterministic plan hash (plan_sha256) and explicit client approval confirmation before spending credits.
Project Defaults:
Model: omni_flash (Omni 1.1 Flash)
Resolution: 360p (native Flow resolution)
Aspect Ratio: 16:9 (landscape)
Duration: 4s per scene
Multi-Scene I2V Batch & Stitch: Full toolset to plan scenes, generate clips, poll completion, and stitch into a final MP4 via ffmpeg.
Related MCP server: google-flow
claude_desktop_config.json){
"mcpServers": {
"yenflow": {
"command": "uvx",
"args": ["yenflow", "mcp", "run"]
}
}
}~/.pi/agent/mcp.json){
"mcpServers": {
"yenflow": {
"command": "yenflow-mcp"
}
}
}python -m yenflow.mcp
# or
yenflow mcp runyenflow serve --transport http --port 8000Tool | Description | State Handling |
| Validates scenes, estimates credit cost, and computes deterministic | Pure/Stateless: returns plan structure and hash directly to client. |
| Generates a single scene end-to-end in silent headless background mode. Returns local MP4 path. | Gated by |
| Submits scene to Flow and returns | Gated by |
| Checks status of a generation by | Stateless: queries Flow directly, returns status to client. |
| Stitches multiple clip paths into a final video with optional audio via ffmpeg. | Concatenates files and validates resolution/duration. |
| Full Veo video generation (t2v, i2v, r2v) with optional character references. | Standalone video generation. |
| Google Imagen text-to-image and image-to-image generation. | Standalone image generation. |
| Query current Google Flow credit balance. | Read-only check. |
| Verify Google Flow session status. | Read-only probe. |
An AI agent orchestrating an Image-to-Video story follows a clean 4-step sequence:
[Agent / MCP Client] [Yenflow MCP Server]
| |
| 1. flow_plan_batch(scenes=[...]) |
|------------------------------------------------------->|
|<-------------------------------------------------------|
| Returns { plan_sha256, estimated_credits, scenes } |
| |
| 2. Review plan_sha256 & confirm approval |
| |
| 3. flow_generate_scene(..., approved=True) |
|------------------------------------------------------->|
|<-------------------------------------------------------|
| Returns { status: "completed", local_path: ... } |
| |
| 4. flow_stitch_video(clip_paths=[...]) |
|------------------------------------------------------->|
|<-------------------------------------------------------|
| Returns { status: "completed", output_path: ... } |Yenflow also includes a complete Click CLI:
# Check Google Flow authentication
yenflow auth status
# Check available credits
yenflow credits
# Run health diagnostics
yenflow doctor
# Start MCP stdio server
yenflow mcp run
# Start MCP HTTP server
yenflow serve --port 8000# Lint & Format
ruff check src tests
ruff format --check src tests
# Static Type Verification
pyright src
# Test Suite
pytest -q tests/auth/test_verification.py tests/test_flow_batch.py tests/mcpMIT License.
Generate a single video scene end-to-end in silent headless background mode. Returns the local path to the generated MP4 file upon completion.
Inputs: prompt (required), start_frame (required), model, aspect, +9 more. Returns: object
Plan an I2V video batch statelessly. Validates scenes and parameters, computes credit cost and deterministic SHA-256 plan hash, and returns the plan state to the MCP client. Zero server-side persistence.
Inputs: scenes (required), count, model, aspect, +4 more. Returns: object
Statelessly check Google Flow for the completion status of a video generation by media ID and project ID.
Inputs: media_id (required), project_id (required), profile. Returns: object
Stitch multiple MP4 video clip files together into a single unified video with optional audio track using ffmpeg.
Inputs: clip_paths (required), output_path (required), audio_path. Returns: object
Submit a single video scene to Google Flow in silent background headless mode (zero window popups, no stolen focus). Returns media_id immediately to the client for tracking.
Inputs: prompt (required), start_frame (required), model, aspect, +7 more. Returns: object
Non-interactive, credit-free Flow session probe (#497). Call this BEFORE a generation tool to fail fast on expired auth — the queue is async, so an auth failure otherwise surfaces only later from the daemon. Never starts an interactive login flow; may boot a short-lived headless browser only if cookie decryption requires the Playwright fallback. May take up to ~45s on a slow network.
Inputs: profile. Returns: object
List the saved Flow CHARACTER entities in a project, with their entity ids. Read-only and spends no credits. Call this to discover what you can attach: an id goes to reference_entities on the generate tools, and a display_name can be used as an @Name mention in a prompt (same wire, they dedupe). Drives a browser session, so it is slower than the catalog tools.
Inputs: project (required), profile. Returns: object
Show one saved Flow CHARACTER entity by id or by exact display name. Read-only and spends no credits. Exactly one of entity_id or name is required; an ambiguous name is an error rather than a guess, which is the reason to prefer the id. Drives a browser session.
Inputs: project (required), name, profile, entity_id. Returns: object
List the preset voices available for a Flow Character's TTS. Static lookup — no network, no browser, no cost. Call it before creating a character to choose a valid voice name.
Inputs: no named parameters. Returns: object
Generate an image using Google Flow's Imagen model. Produces 1-4 images from a text prompt. Models: nano2 (fast), nano-pro (balanced), image4 (highest quality). Aspects: 1:1, 9:16, 16:9, 4:3, 3:4. The prompt supports @AssetName mentions to tag saved project characters/assets by name (resolves to referenceEntities/referenceImages). Reference a SAVED named asset via @Name; reference an arbitrary one-off image via reference_images. See docs/REFERENCE_STRATEGIES.md. On accounts served from flow.google.com, use an existing project and local reference files; UUID/entity references and image4 are not ported to that composer yet and fail before submit; retrying will not clear it. Returns local file paths to the generated images.
Inputs: prompt (required), seed, wait, count, +12 more. Returns: object
Generate a video using Google Flow's Veo model. Modes: t2v (text-to-video), i2v (image-to-video), r2v (reference-to-video). Aspects: 9:16, 16:9. Optional model (veo_lite/veo_fast/veo_quality/omni_flash), duration (seconds), and count select the Veo model, clip length, and batch size (CLI parity). The prompt supports @CharacterName mentions to tag saved project characters by name (resolves to referenceEntities). Reference a SAVED character via @Name; pass one-off ingredient images via reference_images. See docs/REFERENCE_STRATEGIES.md. Optional ui_mode ('classic'/'auto') verifies the classic editor pre-submit and aborts before spending credits if unreachable; 'agentic' is not supported for video. Returns the local file path to the generated video.
Inputs: prompt (required), mode, wait, count, +14 more. Returns: object
Read the current Google Flow credit balance for one saved profile or all profiles. This is read-only and spends no credits. Set all_profiles=true when choosing an account for generation; partial profile failures remain visible in the result.
Inputs: profile, all_profiles. Returns: object
Add a persistent instruction card to a Flow project's Agent-Mode brief (credits-free). Each ref is classified automatically: local image path → uploaded as an image reference; asset UUID → image reference; anything else → character id/name.
Inputs: text (required), title (required), project (required), refs, +2 more. Returns: object
Declaratively FULL-SYNC a Flow project's brief: REPLACES all existing instruction cards with the given set (destructive — cards not listed are removed). Each card is {'title', 'text', 'ref': [...], 'enabled'}. Credits-free.
Inputs: cards (required), project (required), profile. Returns: object
List a Flow project's persistent Agent-Mode instruction cards (reads the live server brief). Credits-free.
Inputs: project (required), profile. Returns: object
Remove one instruction card from a Flow project's brief, selected by title or card id (exactly one). Credits-free.
Inputs: project (required), title, card_id, profile. Returns: object
Enable or disable one instruction card on a Flow project's brief, selected by title or card id (exactly one). Credits-free.
Inputs: enabled (required), project (required), title, card_id, +1 more. Returns: object
Turn a Flow project's brief master switch on or off. When off, NO cards apply even if individually enabled. Cards are left untouched. Credits-free.
Inputs: enabled (required), project (required), profile. Returns: object
List all projects in the local gflow catalog. Returns project IDs, names, and creation dates from the SQLite database.
Inputs: limit, offset, profile. Returns: object
List available gflow prompt tools (name, title, description, category).
Inputs: no named parameters. Returns: object
Alias for flow_generate_scene.
Inputs: prompt (required), start_frame (required), model, aspect, +9 more. Returns: object
Alias for flow_plan_batch.
Inputs: scenes (required), count, model, aspect, +4 more. Returns: object
Alias for flow_poll_scene.
Inputs: media_id (required), project_id (required), profile. Returns: object
Alias for flow_stitch_video.
Inputs: clip_paths (required), output_path (required), audio_path. Returns: object
Alias for flow_submit_scene.
Inputs: prompt (required), start_frame (required), model, aspect, +7 more. Returns: object
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 25 tools
Several tools are exact aliases (yenflow_* for flow_*), creating duplicate entries that an agent must disambiguate. More importantly, gflow_generate_video, flow_generate_scene, and flow_submit_scene all cover video generation but with unclear boundaries, making misselection likely.
Names are mostly snake_case but mix three prefixes—gflow_, flow_, and yenflow_ aliases—without a clear rule. Verb usage is also uneven: 'rm' instead of 'remove', 'get_credits' vs 'list_*', and 'plan_batch'/'submit_scene'/'generate_scene' follow different patterns.
Twenty-five tool names is too many, and five are exact aliases, inflating the count to 25 while only representing about 20 distinct operations. The surface feels heavier than needed, especially with six instruction-card tools and three scene-generation variants.
The set covers a broad workflow: auth, credits, projects, characters, instruction cards, image/video generation, scene planning/submission/polling, and stitching. However, character creation/update/delete is absent despite character voice and listing tools, and batch planning has no accompanying batch submission tool, creating notable workflow gaps.
Google Flow's agent: chat that plans and makes images and videos, approvals, your own media
Plan, compare, price, generate, and recover AI video from compatible MCP clients.
AI video, image and audio generation with 150+ AI models from any MCP agent
AI image, video, voice and music generation over MCP, routed to Veo 3.1, Seedance 2.5 and more.
