What openFlow MCP server does
The openFlow MCP server lets an MCP-compatible agent operate Google Flow for image, video, music, and character-generation workflows. Images use Google’s Nano Banana models, while video generation uses Veo capabilities available through Flow. Results include direct download links, allowing an agent to save generated media into a project.
The server runs work against the user’s Google account. Image generation is described as free on a normal account, subject to the account and service limits. Video uses the account’s Google credits and also requires an openFlow subscription. A new key includes 25 image generations before a GitHub account is linked; linking removes that limit.
How it works
Clients connect to the hosted MCP endpoint at https://openflowmcp.com/mcp over HTTP. After connecting, an agent can register a key when asked to check openFlow status, then guide the user through linking Google. Authentication happens through a script on the user’s machine that opens Google’s own login page. The Google password remains in that browser flow; the service receives a token instead.
The openFlow MCP server provides model-aware operations through tool descriptions intended for agents. list_models returns the currently verified model identifiers and credit prices rather than requiring the client to guess model names. Jobs can be checked while running, and queued work can retry when the account’s simultaneous-generation limit is reached.
Setup and configuration
For Claude Code, add the HTTP server with:
claude mcp add --transport http openflow https://openflowmcp.com/mcp
Other clients can use an MCP server entry with the same URL and an optional bearer key:
{
"mcpServers": {
"openflow": {
"type": "http",
"url": "https://openflowmcp.com/mcp",
"headers": { "Authorization": "Bearer YOUR_KEY" }
}
}
}
After connecting, ask the agent to check openFlow status and connect the Google account. No SDK or separately purchased Google API key is required by the documented setup.
Tools and capabilities
Available tool groups include:
- Images:
generate_image, upscale_image, and generate_scene.
- Video: text-to-video, image-to-video, reference-to-video, frame interpolation, extension, editing, and upscaling.
- Characters: create characters from photos or descriptions, generate facial variations, create turnarounds and expression sheets, and manage saved characters.
- Music: generate a track with lyrics and style controls, then split it into vocal and instrumental stems.
- Status and files: inspect Flow health, dashboard data, credits, jobs, models, and generated media.
Five MCP prompts cover recurring workflows such as character UGC video, turnarounds, expression sheets, generate-then-refine tasks, and multiple camera angles in one location. Reference images can preserve faces, products, rooms, or other approved elements across generations. generate_scene helps maintain object positions and depth, but additional room references may be needed to preserve materials and architecture.
Limitations and notes
A Google account supports up to eight simultaneous generations according to the documented measurements. Nano Banana Pro has a separate daily account quota. Multiple linked Google accounts can distribute work and increase available rate limits.
Files remain on the service until storage pressure requires cleanup; files already downloaded are eligible for removal, while files not yet fetched are not deleted under the stated policy. The documentation also states that tokens are encrypted with AES-256-GCM and that the encryption key is stored separately from the database.
Video lengths documented for the available workflows include 4, 6, 8, and 10 seconds. Image-to-video is better suited to preserving an existing frame, while reference-to-video is intended to preserve identity from an example and may alter small details.