One MCP server for 450+ models, from Veo 3.1, Kling and Seedance to Nano Banana, GPT Image, speech and music. Connect Claude or ChatGPT by signing in, or Cursor, VS Code and Claude Code with a key, and pay per result from one balance.
Their connector screens take an address, not a key. Add this one, sign in to GenMagic when asked (or create a free account) and choose Allow. The app gets a key of its own, and every generation draws on your balance.
Sign-in address
https://genmagic.co/api/mcp/connect
Claude (web, desktop and mobile)
Open Customize › Connectors and select Add custom connector.
Name it GenMagic, paste the sign-in address below as the server URL and select Add. Leave the OAuth fields empty.
Select Connect, sign in to GenMagic and choose Allow.
On Claude's free plan you can add one custom connector. On Team and Enterprise plans an owner adds it for the organization, and each member connects with their own account.
ChatGPT
In Settings, open Apps (called Connectors in some versions), then Advanced settings, and turn on Developer mode.
Create a connector named GenMagic with the sign-in address below and OAuth as its authentication.
Sign in to GenMagic, choose Allow, then pick GenMagic from the + menu in a chat.
Developer mode is offered on some ChatGPT plans and can be turned off by a workspace admin.
Claude Code, Cursor and VS Code
These take either address. With the sign-in address they open GenMagic in your browser the first time; with the key address you send a key you create for free.
Claude Code, signing in (then run /mcp and choose Authenticate)
claude mcp add --transport http genmagic https://genmagic.co/api/mcp/connect
Add to VS CodeAdd to CursorVS Code asks for your key once; in Cursor, replace the placeholder key in its MCP settings.
More apps and tools
These apps add GenMagic from a connector form: give them the sign-in address above, sign in to GenMagic when they ask and choose Allow. No key to copy.
Lovable
Open Connectors, choose Add custom and then MCP server, keep OAuth and select Add & authorize.
Works in Lovable's chat; a published Lovable app cannot call it.
Bolt.new
In the chatbox, + › Connectors › Manage connectors › Custom MCP server, transport HTTP, authentication MCP OAuth.
Turn the connector on in each project that should use it.
v0
In the prompt form, + › MCPs, add a custom server and choose OAuth.
v0 uses the tools while it builds; the generated app does not call them.
Perplexity
Open Plugins, choose Add custom › Add MCP connector › Remote, and under Advanced set authentication to OAuth and transport to Streamable HTTP.
In an organization, an admin must allow members to add custom connectors.
Mistral Vibe (formerly Le Chat)
On the Connectors page, + Add Connector › Custom MCP Connector, then Connect and sign in.
Adding a custom connector is for the account's administrator, which on Free and Pro plans is you.
Notion custom agents
An admin first turns on custom MCP servers; then in the agent, Settings › Tools & Access › Add connection › Custom MCP server.
Notion offers MCP connections on Business and Enterprise plans.
Raycast
Run Install MCP Server, choose transport HTTP and OAuth type Dynamic, then Sign In.
MCP is part of Raycast Pro, on desktop and iOS.
Open WebUI
In Admin Settings › Integrations › External Tool Servers, add a connection of type MCP (Streamable HTTP) with auth OAuth 2.1.
Needs Open WebUI 0.6.31 or newer; enable the tool per chat rather than as a model default.
Zed
In Settings › AI › MCP Servers, choose Add Remote Server with the address and no Authorization header; Zed then asks you to sign in.
With an Authorization header set, Zed sends the key instead of signing in.
Gumloop
In Customize › Connectors, click Add Custom MCP, enter the address in MCP Server URL, Connect, then Authenticate when Gumloop detects the sign-in.
Tools run without an approval prompt, so every generation tool is live at once.
TypingMind
Go to Plugins › MCP Store › Add custom connector, enter the address as the server URL and follow the sign-in.
The Personal version first asks for an MCP Connector (TypingMind Cloud with a free account, or your own).
Jan
In Settings › MCP Servers, + Add MCP Server, transport HTTP, address in Command / URL, turn it on and choose Sign in on its Auth line.
The sign-in wait times out after five minutes; Clear authentication forgets it.
LibreChat
In MCP Settings, Add MCP, transport Streamable HTTPS, authentication OAuth with Client ID and Client Secret left empty.
Needs a role allowed to create MCP servers; an admin can also add it in librechat.yaml.
Sim
In Settings › MCP tools, Add server with a name and the address, save, then Authorize on the server.
Needs Write or Admin access in the workspace.
Goose
Add a remote extension of type Streamable HTTP (goose configure › Add Extension, or Extensions in the desktop app); Goose signs in by itself.
Goose supports Streamable HTTP only, not SSE.
These agent and automation builders take the key address and your API key as a bearer token:
Zapier (MCP Client)
On Apps, + Add connection, choose MCP Client; Server URL the key address, TransportStreamable HTTP, OAuth No, Bearer Token your key. Then use Run Tool in a Zap or the connection in a Zapier agent.
Zapier's MCP Client is in beta. GenMagic also has its own Zapier steps: see the Zapier page.
Make (MCP Client)
Add the MCP Client module, Create a connection, + New MCP Server, the key address in URL and your key in API Key / Access token; then Call a tool.
In open beta at Make. GenMagic's own Make app is on the Make page.
Activepieces
In an agent's Agent Tools, add an MCP server with the key address and an access token; or in a flow, the MCP Client piece's Call Tool with a Bearer Token connection.
The MCP Client piece needs Activepieces 0.92.4 or newer when self-hosted.
n8n AI Agent
Under an AI Agent's Tools, add MCP Client Tool: Endpoint the key address, Server TransportHTTP Streamable, AuthenticationBearer Auth with your key.
For workflows, the verified GenMagic node is simpler: see the n8n page.
Lindy
In Integrations, + Add › MCP, the key address as Server URL, and under Advanced your key as the Authorization bearer token; Connect.
Lindy turns every GenMagic tool into an action; set Guardrails to Ask for approval to confirm each generation.
Langflow
In Settings › MCP Servers, Add MCP Server › HTTP/SSE, the key address, and a header Authorization whose value is a global variable holding Bearer your-key.
The stored value must include the word Bearer.
Flowise
On an Agent node add Custom MCP with the config {"url": "https://genmagic.co/api/mcp", "headers": {"Authorization": "Bearer {{$vars.genmagicKey}}"}}, after creating a variable genmagicKey with your key.
Refresh Available Actions and pick the tools to expose.
Msty Studio
In Toolbox › Tools, Add New Tool › HTTP, the key address as MCP Server URL and a header Authorization: Bearer your-key.
Streamable HTTP only, which is what GenMagic serves.
Command-line agents and config files take the key address with your key as a bearer header (Codex and Gemini CLI can also sign in: point them at the sign-in address and run codex mcp login or /mcp auth).
Every other client config, the raw JSON-RPC calls and each tool's arguments are in the developer docs.
A real session
A product photo and a clip, in three tool calls
Run against the live server on October 29, 2026, with these exact arguments. Nothing was edited afterwards; the costs are what each call was billed: $0.33 in all.
1. generate_image
No model named for the photo, so the default ran: Nano Banana 2.1, in about 20 seconds, for 6.9 cents. The result came back as the image itself plus a hosted address.
{ "prompt": "Product photo of a matte black ceramic pour-over coffee dripper on a light oak counter, morning window light, soft shadows, steam rising from a glass carafe below, minimal Scandinavian kitchen, vertical composition",
"size": "1024x1536" }
2. create_video
The photo's address as the opening image, Veo 3.1 Lite, 4 seconds at 720p with sound. It returned a job id at once and charged nothing yet.
{ "prompt": "Slow push-in. Coffee drips steadily from the dripper into the glass carafe, steam curls up through the morning light. Quiet kitchen ambience, the soft patter of dripping coffee.",
"model": "google/veo-3.1-lite",
"image": "<the photo's hosted URL>",
"duration": 4, "resolution": "720p",
"generate_audio": true, "aspect_ratio": "9:16" }
3. get_video
Polled until the clip was ready, then charged once: 26 cents, which is 4 seconds at Veo 3.1 Lite's price with sound. Every charged call returns the balance left, so the agent always knows what it can still afford.
The photo, Nano Banana 2.1The clip, Veo 3.1 Lite with sound
list_models: List the models you can pick, with ids and prices (filter by category or search); no key needed.
generate_text: Generate text, code, an SVG graphic, or a full web page with any text model (optional type and model).
generate_image: Generate or edit an image with any image model; type=logo for a brand mark.
generate_speech: Text to speech with any speech model and voice, voiced verbatim.
generate_music: Original music track from a description.
create_video: Start an async video job with any text-to-video model, length, resolution and sound, or animate an image (returns an id, charges nothing yet).
get_video: Poll a video job; charges once on completion.
Every generate tool takes an optional model id from the model library, and a model a tool does not run is refused by name, never swapped for another. Results come out on-brand when you turn on personalization in GenMagic. Building workflows instead of agents? The same models are in our n8n node.
Questions
About the GenMagic MCP server
Which apps can use the GenMagic MCP server?
Claude on the web, desktop and mobile and ChatGPT connect by signing in, and so do Lovable, Bolt.new, v0, Perplexity, Mistral Vibe, Notion custom agents, Raycast, Open WebUI, Zed and Goose. Claude Code, Cursor, VS Code, Windsurf, Codex CLI, Gemini CLI, Cline, LM Studio and any client that speaks remote MCP over Streamable HTTP connect with an API key or by signing in. GenMagic is also listed in the official MCP Registry as io.github.yumaheymans/genmagic.
Do I need an API key?
Not for Claude or ChatGPT: signing in creates a key for that app, named after it, which you can see and revoke under API keys on your account page. For clients that take a header, create a free key and send it as Authorization: Bearer. Listing the models and their prices needs no key at all.
What does it cost?
You pay for each result from one prepaid balance, at the price list_models shows, and the cost of every call comes back with its result. A new account gets a one-time $0.50 starter balance once its email is verified, and top-ups start at $5. For an agent that generates every day, a monthly plan adds more credit than it costs (Pro: $20 of credit every month for $18; Studio: $55 of credit every month for $49; Agency: $167 of credit every month for $149). The session on this page cost $0.33 for a product photo and a 4-second clip with sound.
How does video work over MCP?
create_video starts the clip and returns a job id without charging anything. The agent then calls get_video every few seconds until the clip is ready, and the clip is charged once, when it is delivered. Pass an image to animate a still; the clip takes the image's shape unless you name an aspect ratio.
Can the agent pick the model?
Yes. list_models returns every model with its id, price per image or per second, resolutions, clip lengths and sound support, so the agent can choose by price or quality. With no model named, each tool runs a good default (Veo 3.1 Lite for video).
How do I disconnect an app?
Remove the connector in the app, and revoke its key under API keys on your account page. A revoked key stops working at once, and the app asks you to sign in again if you reconnect.