artificer-mcp

Status: Pre-release (v0.x). API may change before v1.0.

artificer-mcp is a public MCP server for AI-powered creative media generation and processing. One server covering the full creative pipeline:

• Image generation via Gemini (Imagen family) + Nano Banana
• Video generation via Veo
• Speech generation via Gemini TTS (prebuilt voices, natural-language style/accent control)
• Music generation via Lyria 3 (batch) + Lyria RealTime (streaming)
• Image processing via ImageMagick (57 tools — resize, composite, text, social cards, ad creatives, app icons, and more)
• Video + audio post-processing via FFmpeg (16 video + 9 audio tools — concatenate, trim, aspect, transitions, overlays, subtitles, b-roll, encoding controls, image-to-video, audio mixing with sidechain ducking, loudness normalization, and more)
• Pluggable storage — local filesystem + Google Cloud Storage out of the box; S3 / OneDrive stubbed for contribution; 7 storage tools (upload, download, list, delete, exists, public URL, signed URL)
• Opinionated workflows — one-call chains for common patterns (brand asset pack, social carousel, multi-element carousel compose, talking-head video, ad creative set, 9:16 short-form video for IG/TikTok/YT/FB, narrated explainer)
• Brand spec — one JSON env var (ARTIFICER_BRAND_SPEC) defines colors, fonts, TTS voice, music prompt, and logo variants so every tool can soft-default to project-consistent choices
• Prompt guides — structured prompt guidance per AI provider (Imagen, Nano Banana, Veo, Gemini TTS, Lyria) with official-doc references + field-tested good/bad examples

Built for the Model Context Protocol. Works with Claude Code, Claude Desktop, Cursor, and any MCP-compatible client.

Repository: github.com/bthurlow/artificer-mcp

Why artificer-mcp

Current creative-media MCP servers are fragmented — one for ImageMagick, one for FFmpeg, one for Veo, one for Nano Banana, each with its own quirks (local-path-only inputs, hardcoded model enums, safety-filter crashes, hanging processes). artificer-mcp consolidates them into one coherent server with URI-first I/O, dynamic model IDs, pluggable providers, and first-class prompt guidance.

Looking for just ImageMagick without AI generation dependencies? Use the focused imagemagick-mcp instead.

Prerequisites

• Node.js 22+
• ImageMagick 7+ with magick in PATH — install
• FFmpeg 6+ with ffmpeg in PATH — install
• Google Gemini API key (if using image or video generation) — get one
• Google Cloud credentials (if using GCS storage) — GOOGLE_APPLICATION_CREDENTIALS env var pointing to a service account JSON

Installation

From source

git clone https://github.com/bthurlow/artificer-mcp.git
cd artificer-mcp
yarn install
yarn build

Claude Code

claude mcp add artificer -- node /path/to/artificer-mcp/dist/index.js

Claude Desktop / Cursor

Add to your MCP config:

{
  "mcpServers": {
    "artificer": {
      "command": "node",
      "args": ["/path/to/artificer-mcp/dist/index.js"],
      "env": {
        "GOOGLE_API_KEY": "your-gemini-api-key",
        "GOOGLE_APPLICATION_CREDENTIALS": "/path/to/gcs-service-account.json"
      }
    }
  }
}

Environment variables

| Variable | Purpose | |---|---| | GOOGLE_API_KEY | Required for all Gemini/Imagen/Veo tools. | | GOOGLE_APPLICATION_CREDENTIALS | Required for gs:// storage URIs. Path to a GCP service-account JSON. | | ARTIFICER_IMAGEN_MODEL | Override default model for gemini_generate_image (fallback: imagen-4.0-generate-001). | | ARTIFICER_IMAGEN_EDIT_MODEL | Override default model for gemini_edit_image (fallback: imagen-3.0-capability-001). | | ARTIFICER_IMAGEN_UPSCALE_MODEL | Override default model for gemini_upscale_image (fallback: imagen-4.0-upscale-preview). | | ARTIFICER_VEO_MODEL | Override default model for gemini_generate_video (fallback: veo-2.0-generate-001). | | ARTIFICER_NANOBANANA_MODEL | Override default model for gemini_nanobanana_generate_image (fallback: gemini-2.5-flash-image). | | ARTIFICER_TTS_MODEL | Override default model for gemini_generate_speech (fallback: gemini-2.5-flash-preview-tts). | | ARTIFICER_LYRIA_MODEL | Override default model for gemini_generate_music (fallback: lyria-3-clip-preview). Also valid: lyria-3-pro-preview. | | ARTIFICER_LYRIA_LIVE_MODEL | Override default model for gemini_generate_music_live (fallback: models/lyria-realtime-exp). | | ARTIFICER_BRAND_SPEC | Optional JSON with project-wide brand defaults (colors, fonts, TTS voice, music prompt, logo variants). All fields optional. See docs/brand-spec.md. |

Per-call model args always win over env overrides.

npx

claude mcp add artificer -- npx artificer-mcp

Architecture

• URI-first API — every tool accepts local paths (./foo.jpg, file:///...) or URIs (gs://bucket/key, https://...). Storage resolution happens inside the tool.
• Per-provider AI tools — no unified generate_video enum; separate tools like gemini_generate_video_from_image. Each surfaces its native params. More providers pluggable.
• Dynamic model IDs — pass any model string. No hardcoded enums; the MCP stays forward-compatible as providers release new models.
• Pluggable storage — StorageProvider interface with URI scheme routing. Local + GCS fully implemented. S3 + OneDrive stubbed for contribution.
• Two-tier tools — primitives (one-op) and opinionated workflows (chained) clearly separated in descriptions.
• Prompt guidance — *_prompt_guide tools return structured markdown with official doc links, prompt templates, good/bad examples, and model-specific notes.

See docs/ for detailed documentation per area.

Documentation

| Doc | Covers | |-----|--------| | docs/storage.md | Storage providers, URI routing, extending with new providers | | docs/image-gen.md | Gemini image generation tools (Imagen + Nano Banana), prompt patterns | | docs/video-gen.md | Veo video generation tools, model selection, prompt patterns | | docs/audio-gen.md | Gemini TTS speech generation + Lyria music generation (batch + realtime) | | docs/video-audio-post.md | FFmpeg-backed video/audio post-processing tools | | docs/workflows.md | Opinionated compound workflows | | docs/brand-spec.md | Project-wide brand defaults via ARTIFICER_BRAND_SPEC | | docs/prompt-guides.md | How prompt guides are structured and extended | | docs/adr/ | Architectural decision records |

Development

yarn install       # Install dependencies
yarn dev           # Run in development (tsx, no build needed)
yarn build         # Build for production (tsup)

# Checks
yarn lint          # ESLint
yarn lint:fix      # ESLint auto-fix
yarn format        # Prettier write
yarn format:check  # Prettier check
yarn typecheck     # tsc --noEmit
yarn test          # Vitest (unit + integration)
yarn test:unit     # Unit tests only
yarn test:integration  # Integration tests (requires ImageMagick + FFmpeg)
yarn test:coverage # Coverage report
yarn ci            # lint + format:check + typecheck + build + test:unit (what CI runs)

Contributing

Contributions welcome, especially for:

• Additional storage providers (AWS S3, OneDrive, Azure Blob, etc.)
• Additional AI generation providers (Runway, Kling, OpenAI, etc.)
• Audio generation providers (ElevenLabs, Suno, etc.)
• Prompt guide improvements and corrections

See docs/ for extension contracts. Before opening a PR, please run yarn ci to make sure all linting, types, build, and tests pass.

License

Apache License 2.0 — see LICENSE.

Copyright 2026 bthurlow.