# Install Vibany in your agent Vibany is an AI image, video and chat platform. Its MCP server lets the user's agent chat, generate images and make videos with the user's own Vibany account: points, history and storage are the same as on the website. You are reading this because the user asked you to connect Vibany. Follow the section for the client you are running in, then tell the user what is left for them to do. ## Server - Name: `vibany` - URL: `https://vibany.com/api/mcp` - Transport: Streamable HTTP (remote server; nothing to install locally) - Auth: OAuth 2.1 sign-in, or an API key sent as `Authorization: Bearer ak_…` ## Rules 1. Prefer the client's own `mcp add` command. When you edit a config file instead, merge the `vibany` entry into the existing file and keep every other server; create the file if it does not exist. 2. Never ask for the user's Vibany password and never sign in for them. OAuth sign-in happens in the user's browser. 3. Use the user-level (global) config unless the user asks for a project-level one, so Vibany works in every project. 4. If a command fails because the client version does not support it, fall back to the config file shown in the same section, or to “Other MCP clients”. 5. Steps marked **User step** happen in the client's UI or the user's browser (slash commands such as `/mcp`, sign-in buttons, consent pages). Do your part (the add command or the config edit), then stop and tell the user exactly what to do next. Call `get_account` only after they confirm they signed in. 6. `ak_…` in commands and configs is a placeholder. Replace it with the user's real key; never run a command that still contains it. 7. If a server named `vibany` already exists, check its URL first: if it matches, skip adding; if not, remove the old entry (same scope) and add it again. 8. Most clients load MCP servers at start-up: after adding, a new session (or a reload) is needed before the tools show up. Say so when you hand over. 9. Reply to the user in their language; this guide is in English only for you. ## API keys Clients marked “API key” need one. Ask the user to create a key at https://vibany.com/ai#connect (sign in, then “Create key”). Prefer that the user pastes it into the config (or an environment variable) themselves; if they paste it to you, use it only in the `Authorization: Bearer ` header, do not print it back, do not commit it to a repository, and keep it in the user-level config rather than a project file. ## Your client ### Claude Code Client id: `claude-code` · Auth: OAuth sign-in in the user's browser (no API key needed) 1. Add Vibany from your terminal: ```bash claude mcp add --transport http --scope user vibany https://vibany.com/api/mcp ``` 2. **User step** (tell the user; don't do it yourself): Sign in (opens your browser), or start a new Claude Code session, type /mcp and pick vibany: ```bash claude mcp login vibany ``` 3. After the user confirms they signed in, verify the connection (see “After installing”). ## After installing 1. If the client only loads MCP servers at start-up, tell the user to restart or reload it (or open a new session). 2. For OAuth clients, tell the user exactly how to start the sign-in (the **User step** above); they finish it in their browser. 3. Once the user says they are signed in, verify by calling the `get_account` tool. It returns the account's points balance and plan. 4. Then suggest a first request, for example: “Use Vibany to make a cover image for this post”. ## Troubleshooting - `ACCOUNT_SETUP_REQUIRED`: the user has never opened the Vibany website with this account. Ask them to open https://vibany.com once while signed in, then retry. - `401` or an auth prompt that keeps coming back: sign in again (OAuth) or check that the API key is complete and not revoked. - `INSUFFICIENT_CREDITS`: the account is out of points; the user can top up on the website. - `RATE_LIMITED` / `DAILY_LIMIT_EXCEEDED`: wait and retry later; the error says when. ## Tools - `get_account`: Show the signed-in user's Vibany account: points balance (the same balance the website shows), whether the user is a paid user, how many image and video tasks are still running versus the pending limits, and how many points agent (MCP) calls have used in the last 24 hours versus the daily MCP limit (limit null means unlimited). - `list_models`: List the models the user can use for one category (chat, image or video) with their prices in Vibany points and parameter options (qualities, aspects, durations, resolutions, aspect ratios, reference image limits). - `estimate_cost`: Quote how many Vibany points a generate_image or generate_video call would cost, using the same pricing as the website. - `generate_image`: Generate images with the user's Vibany account, exactly like the website: same models, same price, same history. - `generate_video`: Start a video generation with the user's Vibany account, exactly like the website: same models, same price, same history. - `get_tasks`: Check image and video tasks of the signed-in user: status (PENDING, PROCESSING, SUCCESS, FAILED, or NOT_FOUND), result URLs when finished, model and creation time. - `list_recent_creations`: List the signed-in user's most recent Vibany images and videos (including ones made on the website), newest first, with status, result URLs, model and creation time. - `chat`: Ask one of Vibany's chat models and get the complete reply, billed to the user's points exactly like the website's chat tool (token-priced models charge by actual tokens, see list_models; pointsCharged is the real charge). More: https://vibany.com/ai