# FlowReel Agent Documentation

FlowReel is an AI video creation product. The Pro Canvas stores workflows as a CreationGraph.

Human documentation:
- https://wovenverse.online/mcp

MCP manifest:
- https://wovenverse.online/api/mcp/manifest

Local MCP server:
- Run with: node mcp/server.mjs
- Required env:
  - FLOWREEL_BASE_URL=https://wovenverse.online
  - FLOWREEL_EMAIL=<user email>
  - FLOWREEL_PASSWORD=<user password>
- Alternative auth:
  - FLOWREEL_COOKIE=session=...

MCP tools:
- flowreel_login: authenticate.
- canvas_list_projects: list projects.
- canvas_create_project: create a blank canvas.
- canvas_get_graph: read CreationGraph.
- canvas_save_graph: save CreationGraph.
- canvas_add_node: add a node.
- canvas_connect_nodes: connect nodes.
- canvas_build_from_brief: create a basic canvas from a natural-language brief.
- canvas_run_graph: execute the canvas. This can consume credits.

Agent safety rules:
- Build and save canvases freely when requested.
- Never call canvas_run_graph unless the user explicitly asks to generate or run.
- Keep credentials outside Git. Use MCP client env vars, local .env files, or a secret manager.

CreationGraph shape:
```json
{
  "nodes": [
    {
      "id": "n_text",
      "kind": "input.text",
      "position": { "x": 80, "y": 120 },
      "params": { "text": "brief or script" },
      "result": {}
    }
  ],
  "edges": [
    {
      "id": "e_text_video",
      "source": "n_text",
      "target": "n_video",
      "sourceHandle": "text",
      "targetHandle": "prompt"
    }
  ],
  "consistency": { "ratio": "9:16" }
}
```

Node kinds:
- input.text
- input.image
- input.video
- input.audio
- gen.t2i
- gen.t2v
- gen.i2v
- gen.tts
- gen.smart_canvas
- gen.stage3d
- compose
- output