For AI agents: MCP server and markdown

Connect the free SuperDuperAI MCP server to Claude, ChatGPT, Cursor or Claude Code: search the AI video prompt library, build Veo 3 prompts, read any page as markdown. No key, no account.

The server lives at https://superduperai.co/mcp. It speaks MCP over Streamable HTTP, needs no key and no sign-in, and only reads: it searches the library, hands back prompts, builds Veo 3 prompts with the same code as the generator (deterministically, no model), and lists the tools and products. It never generates video and never spends money.

Connect it

Claude (claude.ai and the desktop app)

Settings → Connectors → Add custom connector. Name: SuperDuperAI. URL: https://superduperai.co/mcp. Leave the OAuth fields empty. Then enable it in a chat from the tools menu.

Claude Code

claude mcp add --transport http superduperai https://superduperai.co/mcp

The prompts below appear as commands, for example /mcp__superduperai__viral_veo3_video.

ChatGPT

Settings → Apps & Connectors → Advanced settings → turn on Developer mode, then Create a connector with the URL https://superduperai.co/mcp and no authentication. Which plans can add custom connectors is OpenAI’s decision and changes; their help page has the current list.

Cursor, Windsurf, VS Code and other clients

Check it from a terminal

curl -s https://superduperai.co/mcp -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"build_veo3_prompt","arguments":{"idea":"A dog runs into the sea at sunrise","aspect":"9:16"}}}'

Tools

search_library: Search the free SuperDuperAI library of copy-ready prompts and how-tos for AI and short-form video: video-generation prompts (Veo, Kling, Runway, Sora), viral formats, hooks, scripts, editing, captions, b-roll, music and Claude Code skills. Returns ranked entries with their URL and the start of the prompt; call get_prompt for the whole entry. With no query, the filters alone select, newest first.
get_prompt: The whole entry: the copy-ready prompt (placeholders in {braces}), steps, a filled-in example, tips, what it works with, and whether we ran it (verified). Accepts a slug from search_library or a superduperai.co/library/ URL.
build_veo3_prompt: Turn an idea into a Veo 3 prompt in the order the model reads best (subject and action, scene, camera, light, style, sound, dialogue, duration, negatives). Deterministic, no model: quoted lines in the idea become dialogue, empty parts get safe defaults, vague words such as “cinematic” or “4k” are removed. Returns the Veo 3 paragraph, a picture-only version for models without sound, and the JSON fields. Write the idea in English for best results; the dialogue may stay in its own language.
list_tools: The free SuperDuperAI tools for AI video (prompt generators, frame grabber, safe-zone overlay, caption checker, video trimmer and more). Each runs in the browser with no account; returns names, what each does, and the URL.
list_products: Apps and open-source projects by SuperDuperAI with status, platforms, price and links (App Store, site, GitHub, privacy, support).
contact: How to reach the team: e-mail with a ready mailto (subject and draft from topic) and a link where the person books a call. Nothing is sent or booked by this tool.

Prompts

Clients that show MCP prompts list these as ready-made commands:

viral_veo3_video: Make a viral Veo 3 video from my idea. Picks a short-form format that fits the idea, writes the hook and the beats, and turns each beat into a Veo 3 prompt.
short_video_script: Script a short video. A timed script and shot list for a 30-60 second vertical video, built on the library’s script and hook entries.
hook_ideas: Hooks for the first three seconds. Ten opening lines for a short video, each tied to a hook pattern from the library, ranked by how specific they are.

Read the site as markdown, without MCP

Send Accept: text/markdown to the home page or any library, tool, product or refreshed blog page and you get its markdown twin instead of HTML; or append .md to the path, e.g. /library/video-prompt-structure-seven-parts.md.
/llms.txt is the map of the site for models, /llms-full.txt the whole library in one file.
/library.json and /products.json are the indexes as JSON.
/api/search?q=veo+camera+movement answers a question in one request, with the full prompts of the top entries.
Discovery files: MCP server card, agent card, API catalog, agents.txt.

Rules

Everything here is public. Quote it with a link back; the library text is CC BY 4.0. Please do not use it to train models (our robots.txt says ai-train=no). The server allows 120 requests a minute from one address, which no person and no reasonable agent reaches.

What comes next

The video Studio will have sign-in. When it ships, the MCP server gets generation tools that act on your account, and those will ask you to sign in through OAuth. Nothing above changes: the tools on this page stay open, with no key.

Contact

[email protected] · book a call. An agent can also call the contact tool, which returns the same with a ready-to-send e-mail; it sends nothing itself.

Use SuperDuperAI from your AI assistant