// PROMPT LIBRARY

Prompts you can run in your own agent.

Every entry on this page is a verbatim prompt you can paste into Claude, ChatGPT, Cursor, or Hermes — backed by MCP Media Engine. Each one shows the model, the credit cost, the surfaces it works in, and (where available) a verified MED job ID for the example image.

Looking for the API contract? See the docs. Need model + pricing detail? Visit /models.

run in your agent

Three lines and you're set.

Install MCP Media Engine once, then paste any prompt from this library into your agent. The job runs out-of-band and the asset URL comes back when the model is done.

install

Run via npx

Get a key at app.mcpmediaengine.com and set the env var.

npx -y mcp-media-engine

# set MEDIAENGINE_API_KEY in your env
# get a key at https://app.mcpmediaengine.com
paste a prompt

Use any entry below

Each card has a copy button. Drop the prompt into your agent verbatim.

// in your agent:
// "use the MCP media engine to generate an image with this prompt:
// a lighthouse on a rocky coast at dawn, soft watercolor painting, wet on wet, paper texture, gentle washes, painterly, no text, no words, no logos"
pillars

Five pillars, one library.

The starter batch is balanced across five pillars — roughly 2 reach posts per 1 revenue post.

Selfie transform

Selfie transforms — the highest-engagement pillar. img2img into a new medium.

3 entries

Pet transform

Pet portraits — soft watercolor or cartoon. Strong organic reach.

2 entries

Product shot

Product shots — for brands remaking ecommerce photography at scale.

2 entries

Space restyle

Space restyles — before/after room edits, save-bait format.

2 entries

Style art

Style art — pure style recipes with a constant subject.

1 entry

House style — pick one, then LOCK

The art pillar rotation needs one signature house style repeated across every pure-style post. Three candidates below, each generated via MCP Media Engine Default Image. The founder picks one; the daily routine (MED-1193) inherits the lock and the renderer (MED-1190) picks up the palette tokens.

A misty mountain ridge rendered as ASCII-art inside an old green-phosphor CRT terminal, glowing phosphor-green and deep black, monospace composition.
art pillar candidate
mjob_05035795

Phosphor terminal

MED's phosphor-green terminal framing fused with high-craft generated imagery — distinctive, on-brand, retrocomputing. This is the reference-triangle skaigenerated spirit translated to MED's existing visual language.

palette tokens

job: mjob_05035795-1fb9-473c-ada6-4a3e30373fc4

An abstract Bauhaus poster composition with bold circles, triangles, and grids in muted brick red, mustard, navy, and off-white.
art pillar candidate
mjob_da179573

Bauhaus poster

Bold modernist print aesthetic — circles, triangles, and grids in muted Bauhaus colors. Clean, design-forward, immediately recognizable. Closest to the rourke 'craft over tooling' reference point.

palette tokens

job: mjob_da179573-64c0-4988-891f-cd9efff82289

A botanical still life of a monstera leaf and glass vase, risograph duotone teal and coral with intentional misregistration grain, layered translucent inks.
art pillar candidate
mjob_5f1c549e

Riso grain

Risograph duotone aesthetic with intentional misregistration grain. Warm, hand-printed, indie-zine feel. Highest 'craft' signal of the three candidates.

palette tokens

job: mjob_5f1c549e-036a-4974-a0c1-223ce27b4834

3-step card

Same shape on every entry.

Each prompt page has the same 3-step setup card. The numbers don't change — only the prompt and the model vary.

  1. 01

    Install + key

    npx -y mcp-media-engine and an API key from app.mcpmediaengine.com.

  2. 02

    Attach your photo (if img2img)

    Attach your selfie, pet, or product photo to your agent chat — or paste any image URL.

  3. 03

    Paste the prompt

    Copy the verbatim prompt from the entry page and paste it. The agent submits the job and returns the asset URL when it's done.

works in

Same prompt, four agents.

Every prompt on this page is designed to run in any of these MCP-compatible agents. No per-agent fork needed.

  • Claude Desktop (native MCP)
  • ChatGPT (via MCP / REST)
  • Cursor (MCP settings)
  • Hermes (any MCP-compatible agent)
  • Direct REST API (HTTP, no MCP required)

Policy: use your own photos — don't upload photos of other people without their consent. No celebrities, no third-party faces in our own posts. Every example image on this page is generated via MED itself; job IDs are listed on each entry page.

Ready to run a prompt in your own agent?