# VicSee MCP Server Generate, edit, and upscale AI **video & images** from any agent — [VicSee](https://vicsee.com) as a set of [MCP](https://modelcontextprotocol.io) tools. Works with **Hermes Agent, Claude (Desktop / Code), Cursor, OpenClaw**, and any MCP-compatible client. ## What your agent can do | Tool | What it does | |------|--------------| | `vicsee_list_models` | List available models (Seedance, Veo, Kling, FLUX, Nano Banana, …) + their credit costs | | `vicsee_estimate` | Same arguments as `vicsee_generate`; returns the **exact credits** it will charge and creates nothing | | `vicsee_generate` | Text/image → **video or image** with a chosen model | | `vicsee_get_task` | Poll a task and get the finished media URL (`result.expires_at` — URLs are kept 7 days; download what you want to keep) | | `vicsee_upscale_image` | Upscale an image | | `vicsee_upscale_video` | Upscale a video | | `vicsee_get_credits` | Check your credit balance | Generation is **asynchronous**: `vicsee_generate` (or `vicsee_upscale_*`) returns a task `id` immediately — your agent then polls `vicsee_get_task` until `status` is `completed` and reads the URL from `result.url`. ## Setup 1. Get an API key at **[vicsee.com](https://vicsee.com) → Settings → API** (starts with `sk-`). API access requires a paid plan or credit pack. 2. Add the server to your MCP client config with that key: ```json { "mcpServers": { "vicsee": { "command": "npx", "args": ["-y", "@vicsee/mcp-server"], "env": { "VICSEE_API_KEY": "sk-your-key-here" } } } } ``` That's the whole setup — drop in your key, and your agent can generate. ### Client config locations - **Claude Desktop:** `claude_desktop_config.json` → `mcpServers` - **Cursor:** Settings → MCP → add server (same `command`/`args`/`env`) - **Hermes Agent / OpenClaw:** add to the MCP servers section of your agent config - **Claude Code:** `claude mcp add vicsee -e VICSEE_API_KEY=sk-... -- npx -y @vicsee/mcp-server` ### Optional env - `VICSEE_BASE_URL` — override the API base URL (defaults to `https://vicsee.com/api/v1`). ## Local development ```bash pnpm install pnpm build VICSEE_API_KEY=sk-... node dist/index.js # stdio server # or run from source: VICSEE_API_KEY=sk-... pnpm dev ``` Point your MCP client at the local build by using `"command": "node", "args": ["/abs/path/to/dist/index.js"]`. ## Example agent flow > "Make me a 5-second video of a kitten chasing a laser." 1. `vicsee_list_models` (type: video) → pick e.g. `seedance-2-5-text-to-video` (up to 30s) or `seedance-2-0-text-to-video` (adds 1080p/4K) 2. `vicsee_generate` (model, prompt, duration: 5) → `{ id, status: "pending" }` 3. `vicsee_get_task` (id) … poll … → `{ status: "completed", result: { url: "https://cdn.vicsee.com/…" } }` ## Notes - Result URLs are served from `cdn.vicsee.com` — stable VicSee CDN links. - Each generation costs credits; see `vicsee_list_models` for per-model costs and `vicsee_get_credits` for your balance. ## License MIT