flow-mcp

What problem does it solve? Driving AI image and video generation from outside the SEOSONA Flow Chrome extension normally requires manual prompting in the browser. This Skill lets an agent or the Video AI V2 pipeline call Flow through a local MCP server to generate visual assets programmatically, with idempotency, structured errors, and file export. ## Core Features & Use Cases - 20 MCP tools with FlowResult envelopes: discovery (list_capabilities, list_models, list_voices), generation (gen_image, gen_video, run_workflow), export (export_asset), control (health, cancel_job), and project/memory tools, each returning structured content validated by outputSchema. - Idempotent generation: pass a client_ref (e.g., scene id) so repeated calls return cached assets instead of consuming quota again, surviving server restarts. - Silent b-roll for V2 pipelines: generate videos without baked-in voice so the V2 pipeline can add TTS, captions, and muxing locally, with script.json contract validation between Flow and V2. - Use Case: A video production agent checks health, lists capabilities, generates one silent b-roll clip per scene with client_ref set to the scene id, exports each video to disk, then hands assets to the V2 pipeline for voiceover and editing. ## Quick Start Ask the agent to check Flow MCP health and list capabilities, then generate a silent b-roll video for a scene and export it to disk.