Nsketch MCP
Generate images, video & audio inside your AI agent
https://nsketch.ai/api/mcpAgents
Copy the Nsketch URL
Click the copy button next. You'll need it in the next step.
Open Settings → Connectors
Add a custom connector, name it Nsketch and paste the URL.
Connect and sign in
Click Add → Connect, sign in with your Nsketch account — you're all set, now just ask Claude to generate an image.
Using a terminal agent? It's better to use the · GitHub
Got any questions left?
We've answered the most frequently asked questions
Two ways: a hosted MCP server (nsketch.ai/api/mcp) that connector-based agents like Claude, ChatGPT, and Cursor talk to directly, and the Nsketch CLI (@nsketch/cli) that terminal agents like Claude Code and Codex drive through the installable skills. Both use the same account, credits, and models as the web app.
Claude (claude.ai connectors), Claude Code, Cursor, ChatGPT, and Codex — plus anything else that speaks MCP over HTTP or can run a CLI.
Everything in the Nsketch catalog — 39 image models (Imagen 4, Nano Banana Pro, GPT Image 2, Seedream, Flux…), 49 video models (Veo 3.1, Kling, Sora 2, Seedance, Wan…), and voice models. Run `nsketch models list` or call the list_models tool for the live catalog.
Images and edits, text/image-to-video, voice generation and cloning, lipsync and talking photos, motion transfer, reshoots, upscaling, and background removal — every generation surface of the Nsketch app.
No. You sign in with your Nsketch account through OAuth — the browser opens once, you approve, done. No keys to copy or rotate.
Generations use your Nsketch credits — the same balance as the web app, at the same rates. Your agent can check the balance any time with the get_credits tool or `nsketch account status`.
Images usually finish in seconds. Videos and motion transfer take one to several minutes; agents poll automatically (or the CLI blocks with --wait) and return the result URL when it's ready.
Yes — every result is a hosted URL, so you can feed an image into video generation, upscaling, lipsync, or edits. Local files work too: the CLI and the upload_media tool host them for you.