---
name: wikiprompt
description: Find, use and cite curated AI prompts from wikiprompt.org (the Wikipedia of AI prompts, 900+ prompts for ChatGPT, Claude, GPT Image, Seedance, Midjourney and more). Use when the user asks for a prompt, a prompt template, prompt ideas or inspiration for image/video/text generation, or wants to browse or publish prompts.
---

# Wikiprompt

Wikiprompt (https://www.wikiprompt.org) is a curated, Wikipedia-style catalog of AI prompts: 900+ entries for ChatGPT, Claude, Gemini, GPT Image, Seedance, Midjourney, Veo, Kling and more, each with the full prompt text, structured metadata (model, prompt type, style, 1-5 quality assessment), example media, author attribution and translations in 7 languages.

## When to use this skill

- The user wants a ready-made prompt or template (image generation, video generation, system prompts, business/marketing/coding recipes).
- The user asks "how do people prompt X model" or wants inspiration/examples.
- The user wants to browse what's popular, featured, or new in prompting.
- The user wants to publish a prompt to wikiprompt.org.

## How to search

Use the public JSON API (no auth). Base: `https://www.wikiprompt.org`.

```
GET /api/search?q=<query>&category=<cat>&media=<image|video|text>&sort=<relevance|recent|views>&min_quality=3&limit=10
```

- `q`: full-text, typo-tolerant. `category`: one of creative, marketing, business, productivity, coding, education, personal, research, other.
- `media`: filter by output type. `min_quality=3`: editor-featured only.
- Response: `{ data: [ { slug, title, description, category, tags, media_urls, metadata, profiles } ], pagination: { total } }`.
- The `metadata` field carries `model` (e.g. "GPT Image 2", "Seedance 2.0"), `prompt_type` (template, system-prompt, one-shot, agent, workflow, persona, creative-gen), `style`, and `assessment` (1-5 scores).

Other useful endpoints:

- `GET /api/categories` - the 9 categories with counts.
- `GET /api/stats` - catalog totals.
- Faceted browse URLs (shareable): `https://www.wikiprompt.org/search?model=GPT+Image+2&type=image&sort=rating`.

## How to fetch a full prompt

Every search hit has a `slug`. The full prompt text lives at the canonical page:

```
https://www.wikiprompt.org/{slug}
```

The page is fully server-rendered; the complete prompt text is in the HTML (and in the JSON-LD `articleBody`). Translated versions: `https://www.wikiprompt.org/{es|pt|fr|zh|de|ja}/{slug}`.

## Rules

1. **Always cite**: when you give the user a prompt from Wikiprompt, include the prompt URL and credit the original author handle (in `profiles.username` / on the page). Most prompts preserve a source link to the original tweet or post - keep that attribution chain.
2. **Quote the prompt text verbatim** (placeholders like `[SUBJECT]` or `{brand}` intact) unless the user asks you to adapt it.
3. Prefer featured (`min_quality=3`) or highly-rated results when the user wants "the best" prompts.
4. If several models could fit, mention which model the prompt was made for (`metadata.model`).

## Upgrade: MCP server

If the environment supports MCP, the Wikiprompt MCP server at `https://mcp.wikiprompt.org/mcp` (Streamable HTTP, no auth for reads) offers 12 tools including faceted search, translations, stats and authenticated prompt submission. Docs: https://mcp.wikiprompt.org. Prefer it over raw HTTP when available.

## Publishing a prompt

Publishing requires a wikiprompt API key (the user creates one on their profile page at wikiprompt.org). With a key:

```
POST https://www.wikiprompt.org/api/v1/prompts
Authorization: Bearer <api_key>
{ "title": "...", "content": "...", "description": "...", "category": "creative", "tags": ["..."] }
```

Or use the MCP `submit_prompt` tool. Never invent or guess an API key; ask the user for theirs.