Which integrations are available for this server?

Enables syncing generated images and clips to a Dropbox folder for sharing and backup. Provides access to Google's generative AI models (Gemini, Imagen, Veo, Nano Banana) for image and video generation, as well as vision/OCR capabilities. Enables syncing generated images and clips to an iCloud folder for review on mobile devices. Allows exporting finalized prompts as markdown notes directly into a Logseq vault for knowledge management. Allows exporting finalized prompts as markdown notes directly into an Obsidian vault for knowledge management. Optional integration for additional video generation models via the Replicate API.

How do I use macro-pickle?

1. Click on "Deploy Server". 2. Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state. 3. In the chat, type @ followed by the MCP server name and your instructions, e.g., "@macro-pickle generate a brand-aware image using the 'tech' profile" That's it! The server will respond to your query, and you can continue using it as needed. Here is a step-by-step guide with screenshots.

macro-pickle

by MrJPTech

Overview Schema Source Related Servers Discussions Admin

TypeScript

Local

🥒 macro-pickle

🥒 Local, database-free AI creative tooling — image & video generation plus a typed, brand-aware Prompt Engine, driven through Claude. No database, no web app. Big dill. 🌶️

👀 What it is

macro-pickle is a command-line + MCP toolkit for generating brand-aware imagery and video and for engineering the prompts behind them. There is no database and no web app — every rail writes files to disk, and finalized prompts are exported as copy-paste markdown you can drop into any notes vault.

It runs through Claude — CLI scripts via tsx, a desktop MCP server, and Claude Code skills/commands — with Google (Gemini vision, Imagen, Veo, Nano Banana) and fal.ai (FLUX et al., Kling video) as the generation backends.

Related MCP server: brand-gen

🚦 Quick start (0 to pickle in 60 seconds 🏃💨)

Prerequisites: Node 20+, pnpm, and at least one API key.

git clone https://github.com/MrJPTech/macro-pickle.git
cd macro-pickle
pnpm install

cp .env.example .env.local        # then add your keys (see below)
pnpm lint                         # tsc --noEmit — verifies the install
pnpm prompt --brands              # list the installed brand profiles
pnpm img "a neon pickle mascot, dark studio bg"     # Imagen → PNG on disk

🔑 Keys

Everything is optional except the backend you actually use — nothing is required to build prompts, only to render pixels.

Variable	Needed for	Get one
`GOOGLE_API_KEY`	Gemini vision/OCR, Imagen, Veo, Nano Banana, Omni Flash	ai.google.dev
`FAL_KEY`	Kling video + FLUX / Recraft / Ideogram / SD3.5 images	fal.ai
`BYTEPLUS_API_KEY` · `RUNWAY_API_KEY` · `REPLICATE_API_TOKEN`	Optional extra video providers via the model registry	see `.env.example`

📁 Where output goes

All paths are yours to set — the defaults are all repo-relative, so a fresh clone works with no configuration.

Variable	Default	What it controls
`MACRO_PICKLE_EXPORT_DIR`	`./generated-images`	Where rendered images/clips land. Point it at a synced folder (iCloud/Drive/Dropbox) to review on your phone.
`MACRO_PICKLE_PROMPT_VAULT`	`./exported-prompts`	Where `--save` writes finalized prompt notes. Point it at an Obsidian/Logseq vault to collect them there.
`MACRO_PICKLE_OMNI_DIR`	falls back to export dir	Default output for `pnpm omni` video edits.
`MACRO_PICKLE_BRANDS_DIR`	`content/brands`	Where brand profiles are loaded from — keep private brands outside the repo.

Command	What it does
`pnpm prompt`	Build image/video prompts via the Prompt Engine (`--image`/`--video`, `--brand`, `--json`, `--save`, `--gen`)
`pnpm img`	Imagen 4.0 reference-image generation → `MACRO_PICKLE_EXPORT_DIR`
`pnpm fal`	fal.ai image models (FLUX / Recraft / Ideogram / SD3.5) — Prompt Engine parity
`pnpm nano`	Nano Banana face-lock generation from reference photos (likenesses you have rights to)
`pnpm veo`	Google Veo video — text-to-video, image-to-video, and `--refs` ASSET identity lock
`pnpm kling`	Kling video via fal.ai — i2v / t2v / start+end interpolation
`pnpm describe`	Gemini vision: product OCR + in-use scene recommendation (UGC rail); `--paddle` for the local OCR sidecar
`pnpm ocr`	Optional local PaddleOCR sidecar (high-recall small/CJK text + geometry)
`pnpm analyze-video`	Describe a reference/competitor clip for mirroring
`pnpm scene:new`/`refs`/`select`/`analyze`/`video`	Media pipeline — two-stage: text → reference frames → curate → re-prompt to video (docs/media-pipeline.md)
`pnpm rank-skus` · `pnpm gen-clips`	Store batch clips — rank a store's SKUs by sales → generic reference→video batch (Veo free + Kling paid), spend-gated
`pnpm omni` · `pnpm seedance` · `pnpm wan`	Gemini Omni Flash conversational video editing · Seedance · Wan video rails
`pnpm models`	Browse / select across the cross-provider video-model registry
`pnpm mcp:image`	Run the `macro-pickle-images` desktop MCP server
`pnpm lint`	Type-check the whole toolkit (`tsc --noEmit`)

scripts/lib/prompts/ turns the methodology in content/knowledge/PROMPT-COOKBOOK.md (synthesized from a fleet of reference repos) into typed, deterministic builders — no DB required:

Piece	What it does
`buildImagePrompt`	Nano Banana "Perfect Prompt" formula — Subject + Action + Context + Composition + Lighting + Style
`buildVideoPrompt`	Seedance/Veo "Director Brief" — Scene · Subject · Camera · Action · Audio · Pacing + time-segmented beats + on-screen captions
Brand profiles	`content/brands/*.json` — style, palette, prefer/avoid, suffix, IP-safe cast proxies; auto-applied via `--brand`
UGC layer	`scripts/lib/prompts/ugc.ts` — faceless short-form selling formats (hero still → lifestyle still → 9:16 spot)
`lintScene`	Continuity & IP-safety linter — single-beat, fixed-camera, stationary-cycle, atmospherics

📓 Export convention: finalized prompts are written as copy-paste markdown — pnpm prompt … --save, or save: true on the MCP build tools. They land in ./exported-prompts/ unless you point MACRO_PICKLE_PROMPT_VAULT at a notes vault.

The two-stage workflow at the heart of the toolkit: turn an idea into reference images, curate the winners, then re-prompt them into video — across Nano Banana / Imagen / Veo / Kling / Gemini. A local scene.json manifest tracks every prompt, frame, and clip (full walkthrough).

pnpm scene:new    my-scene        # scaffold scene.json (reference + video prompts)
pnpm scene:refs   my-scene        # Stage 1 — generate reference-frame candidates
pnpm scene:select my-scene <ids>  # curate the keepers
pnpm scene:analyze my-scene       # (optional) Gemini grounding of the picks
pnpm scene:video  my-scene        # Stage 2 — selected frames → Veo clip

Piece	What it does
`scene-store`	Local JSON manifest — prompts, frames, selections, clips per scene
`PromptEnhancer`	Gemini chain-of-thought prompt rewriter (+ exemplar banks) that directs your idea
`groundFrames`	Image → understanding feedback loop — keeps Stage 2 on-subject / on-brand
Veo modes	`ingredients` (ASSET refs) · `firstLast` (first→last frame) · `firstFrame` (i2v)

Built on the same scripts/lib/ clients as the rest of the toolkit — one set of model clients, no duplication.

MCP (macro-pickle-images, pnpm mcp:image) — exposes build_image_prompt, build_video_prompt, describe_product, build_ugc_spot, analyze_reference_video, lint_scene, generate_image, generate_video to Claude Desktop.
Skills (.claude/skills/) — openmontage-video-prompting (cinematography), ugc-shortform-prompting (selling layer), character-lock (identity / face lock).
Commands (.claude/commands/) — /pickle-prompt · /pickle-ugc · /pickle-pipeline (scene:*) · /pickle-winners → /pickle-clips (rank winners → batch clips) · /pickle-ref (clean references) · /pickle-describe (OCR + scene rec) · /pickle-character (face-lock) · /pickle-brand (new brand profile).

Layer	Technology
Language / runtime	TypeScript (strict, ESM) on Node via `tsx`
Image / video	`@google/genai` (Gemini · Imagen · Veo · Nano Banana) · `@fal-ai/client` (FLUX · Kling)
MCP	`@modelcontextprotocol/sdk`
Validation	`zod`
Optional OCR sidecar	Python + PaddleOCR (`scripts/py/`, opt-in)

macro-pickle/
├── scripts/
│   ├── lib/                # generation cores: imagen, fal, veo, kling, nano-banana, vision, paddleocr
│   │   ├── prompts/        # the Prompt Engine (builders, brand, scene, export, ugc, lint)
│   │   └── pipeline/       # scene-store + prompt-enhancer + exemplars + presets
│   ├── pipeline/           # scene:new/refs/select/analyze/video CLI
│   ├── generate-*.ts       # pnpm img / fal / nano / veo / kling
│   ├── build-prompt.ts     # pnpm prompt
│   ├── describe-product.ts · ocr.ts · analyze-video.ts   # vision / OCR rails
│   ├── gen-clips.ts · rank-skus.ts   # store batch-clip rails
│   └── py/                 # optional PaddleOCR sidecar
├── mcp/image-server/       # macro-pickle-images desktop MCP
├── content/
│   ├── brands/             # brand profiles (*.json) — `quiet-desk` is the worked example
│   ├── briefs/             # scene briefs — `example-logo/` shows the format
│   ├── clip-scenes/        # per-store batch-clip scene configs (`example-store.json`)
│   └── knowledge/          # PROMPT-COOKBOOK.md methodology
└── .claude/                # skills + slash commands

📜 License

MIT — see LICENSE. Free as a pickle at a deli counter. 🥪

🧑‍🍳 Lovingly brined by MrJPTech · a PRSMTECH project

📅 Last Updated: July 2026 · Status: 🎨 Local database-free creative toolkit — image + video gen, Prompt Engine, MCP · Vibe: kind of a big dill 🥒

This server cannot be deployed

One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

Maintenance

ActivitySlowing

ResponsivenessNo issues

Resources

GitHub Repository

Need Help?

Related Servers

Related MCP Connectors

distribea-mcp A
io.github.Distribea
Generate images, video, music and voice from your CLI or AI agent. On-brand AI media toolkit.
AI & Machine Learning Image & Video Processing Marketing
21
mcpOAuth
io.styleforge
Brand-aware creative studio for Claude: 200+ tools for on-brand ads, video, email and campaigns.
Marketing Content Management Systems AI & Machine Learning
1
Creative ClawOAuth
co.creativeclaw
On-brand creative studio for AI agents: images, video, audio, and 3D.
Image & Video Processing Multimedia Processing Design Tools
22
NativeAds AOAuth
ai.nativeads.mcp
On-brand ad creative generation: teach it your brand once, generate images and video forever.
AI & Machine Learning Marketing
1

Related MCP Servers

local_ai_gen
AI & Machine Learning Multimedia Processing Audio Processing
davidemodolo
A
license
Not graded
quality
D
maintenance
Local AI generation for images, audio, speech, and 3D models using open source models.
Updated 6 months ago (2026-03-28 18:01 UTC)
3
GPL 3.0
brand-gen
Marketing Design Tools Autonomous Agents
velinussage
A
license
Not graded
quality
D
maintenance
A memory-backed brand generation runtime for agent-led creative iteration. Enables AI agents to plan, generate, review, and improve brand materials with persistent brand memory and structured workflows.
Updated 4 months ago (2026-05-22 11:32 UTC)
MIT
LemGen AI Design MCP
Image & Video Processing AI & Machine Learning Multimedia Processing
gptimage2prompts
A
license
A
quality
C
maintenance
Open-source MCP server for AI image and video creation, enabling prompt library search, prompt enhancement, and media generation from Claude Code, Cursor, and other MCP hosts.
Updated 2 months ago (2026-07-01 13:34 UTC)
8
MIT
imagethis
AI & Machine Learning Image & Video Processing
dbbuilder-org
A
license
Not graded
quality
C
maintenance
One-step AI image generation for Claude Code, allowing users to generate images from natural language prompts with support for multiple backends like Replicate and OpenAI.
Updated 2 months ago (2026-07-08 07:47 UTC)
MIT