# MCP Media Engine

> Async AI media generation API and Model Context Protocol (MCP) server. Generates images, short-form video, and audio for AI agents and production pipelines. Free tier, no credit card.

## What it is
MCP Media Engine is an async API and native MCP server from SlowMotionMagic.com. Agents and applications submit a media generation job to a queue, then receive the finished asset URL via webhook or polling. Exposes intent-level tools (hero image, social teaser, email header, teaser video, article narration) plus generic generate_image / generate_video / generate_audio tools and job lifecycle tools.

## Capability summary
- **Media modes:** image, video, audio
- **MCP transport:** stdio (npm shim) + streamable HTTP (hosted at https://mcp.mcpmediaengine.com/mcp)
- **Execution model:** async job queue (submit -> queued -> processing -> done | failed)
- **Delivery:** webhook callback or polling
- **Batch:** up to 500 jobs per request
- **Authentication:** API key (`me_live_...` or `me_test_...`)
- **Free tier:** 25 credits on signup, no credit card

## Remote MCP install (Smithery, claude.ai, ChatGPT)
- URL: `https://mcp.mcpmediaengine.com/mcp`
- Transport: MCP streamable HTTP (POST + optional SSE stream)
- Auth: `Authorization: Bearer me_live_…` on every request (per-user, no shared secret)
- Health: `GET /health` on the same host returns `{ "status": "ok", ... }`

## MCP tools (intent-specific)
- `generate_article_hero_image` — Hero banner for a blog post or article from title and summary
- `generate_social_teaser_image` — Social media teaser (LinkedIn, Twitter/X, Instagram, Substack)
- `generate_email_header_image` — Wide-format header for marketing emails and newsletters
- `generate_article_teaser_video` — 15–30 second teaser video from article summary
- `generate_article_narration` — Audio narration via text-to-speech

## MCP tools (generic)
- `generate_image` — Text-to-image with provider selection
- `generate_video` — Text-to-video with provider-specific parameters
- `generate_audio` — Text-to-speech synthesis (multiple voices, formats, speeds)
- `wait_for_job` — Block until a job completes and return the final asset URL
- `get_job_status` — Check current state of a job (queued, processing, done, failed)
- `list_jobs` — List recent media generation jobs with filtering

## Compatible agents
- Claude Desktop (native MCP)
- Claude Code (SDK / `.claude/claude.md`)
- Cursor (MCP settings)
- Windsurf (AI settings -> MCP Servers)
- Cline (`cline_config.json`)
- Codex CLI
- Gemini CLI
- Goose
- Any MCP client over stdio
- Direct REST API (HTTP, no MCP required)

## Quick start
1. Get an API key at https://app.mcpmediaengine.com
2. Install the SDK: `npm install @mediaengine/client`
3. Generate a media asset

```typescript
import { MediaEngineClient } from '@mediaengine/client';

const client = new MediaEngineClient(process.env.MEDIA_ENGINE_API_KEY!);

const job = await client.generateImage({
  prompt: 'A professional blog hero image',
  provider: 'the MediaEngine generation pipeline',
});

const result = await client.waitForJob(job.mediaJobId);
console.log(result.outputUrl);
```

## Pricing
- **Free:** 25 credits (one-time on signup)
- **Starter:** $29 / month, 400 credits
- **Pro:** $99 / month, 1,400 credits
- **Enterprise:** Custom pricing, custom credit volume, SLA

## Prompt library (/prompts)
A starter catalog of 10 prompts across 5 content pillars — selfie, pets, product, spaces, and style-art. Each entry has the verbatim prompt, the model + credit cost, the surfaces it works in (Claude, ChatGPT, Cursor, Hermes), a 3-step setup card, and (where available) a verified MED job ID for the example image.

Entry index: https://mcpmediaengine.com/prompts
Per-prompt pages are at https://mcpmediaengine.com/prompts/{slug}.

Pillars:
- **Selfie transforms** — img2img into a new medium (watercolor, pencil sketch, renaissance oil).
- **Pets** — pet portraits in soft watercolor or cartoon styles.
- **Product shots** — clean studio + lifestyle variants for ecommerce brands.
- **Spaces / rooms** — before/after interior restyles (mid-century, scandinavian).
- **Style art** — pure style recipes with a constant subject (e.g. watercolor coastline).

Every entry works in Claude, ChatGPT, Cursor, and Hermes via the MCP server. No per-agent fork. The Default Image model (1 credit) is the recommended starting point; bump to the Premium Image model (4 credits) for text-in-image or strict prompt adherence.

## Image style recipes (verified live — see /docs/styles)
Style is controlled by a short fragment appended to the subject line. Model and aspect_ratio are passed in `inputParams`. Bake exclusions into the prompt text (the Default Image model does not honor a separate `negative_prompt` field).

Subject constant used for verification: "a lighthouse on a rocky coast at dawn".

- **Photorealistic** — `photorealistic, dramatic natural light, shallow depth of field` (Default Image, 1cr, 16:9)
- **Black & White Archival** — `black and white archival photograph, 1930s documentary style, film grain, high contrast` (Default Image, 1cr, 4:5)
- **Pencil Sketch** — `loose pencil sketch, hand-drawn crosshatching, monochrome, sketchbook page` (Default Image, 1cr, 4:5)
- **Cartoon Illustration** — `playful cartoon illustration, bold outlines, flat vibrant colors` (Default Image, 1cr, 16:9)
- **Watercolor** — `soft watercolor painting, wet on wet, paper texture, gentle washes, painterly` (Default Image, 1cr, 4:5)
- **Flat Vector / Minimalist** — `flat vector illustration, minimalist, clean geometric shapes, limited palette, modern poster style` (Default Image, 1cr, 16:9)
- **Isometric 3D** — `isometric 3D render, soft studio lighting, low-poly, clay material, cozy diorama` (Default Image, 1cr, 16:9)
- **Cinematic Film Still** — `cinematic film still, anamorphic lens, teal and orange color grade, atmospheric haze, wide shot` (Default Image, 1cr, 21:9)
- **Retro Travel Poster** — `vintage travel poster, 1950s offset print, bold flat color blocks, screen-print halftone, heroic composition` (Default Image, 1cr, 2:3)
- **Line Art** — `single continuous line art, minimalist drawing, elegant pen on white, unbroken contour` (Default Image, 1cr, 1:1)

Constraints baked into every recipe: `no text, no words, no logos`.

## aspect_ratio values (verified live against production)
`1:1`, `16:9`, `9:16`, `21:9`, `9:21`, `3:2`, `2:3`, `5:4`, `4:5`, `4:3`, `3:4`.

## Model selection
- **Default Image** — 1 credit. Honors `prompt`, `aspect_ratio`, `cfg_scale` (0–15), `steps`, `num_images`, `super_resolution`, `input_image_url` + `denoising_strength` (img2img), `style`, `color_grading`, `film_grain`, `face_swap`, `inpaint_faces`, `tune_id`. Negative prompts are NOT honored — bake exclusions into the prompt text.
- **Premium Image** — 4 credits. Use for text-in-image, strict prompt adherence, and native `negative_prompt` support. Fields: `prompt`, `quality`, `num_images`, `aspect_ratio`, `negative_prompt`, `cfg_scale`, `steps`.

Live catalog: `GET https://api.mcpmediaengine.com/api/models`.

## Links
- Website: https://mcpmediaengine.com
- Documentation: https://mcpmediaengine.com/docs
- API reference: https://mcpmediaengine.com/docs/api
- Style & prompting guide: https://mcpmediaengine.com/docs/styles
- Prompt library: https://mcpmediaengine.com/prompts
- MCP page: https://mcpmediaengine.com/mcp
- Dashboard / API keys: https://app.mcpmediaengine.com
- API base URL: https://api.mcpmediaengine.com
- npm: https://www.npmjs.com/package/mcp-media-engine
- GitHub: https://github.com/Tr1ckyMag1ca1/MediaEngine

## Company
- Publisher: MediaEngine (SlowMotionMagic.com)
- Contact: hello@mcpmediaengine.com
- Support: support@mcpmediaengine.com

See `llms-full.txt` for the complete tool reference, API surface, webhook signing, and job lifecycle.