生成式媒体 MCP 服务器

ADK Rust MCP Toolkit

Production-ready Model Context Protocol (MCP) servers for generative media, built in Rust. Generate images, videos, music, and speech through a unified, provider-agnostic interface.

Features

• 🖼️ Image Generation — Text-to-image with Imagen 3 and Gemini multimodal models
• 🎬 Video Generation — Text-to-video, image-to-video, video extension with Veo 3.1
• 🎵 Music Generation — Full songs with Lyria 3 Pro/Clip, real-time streaming with Lyria RealTime
• 🗣️ Speech Synthesis — High-quality TTS with Chirp3-HD and Gemini voices (30 voices, 70+ languages)
• 🎛️ Media Processing — FFmpeg-powered audio/video manipulation
• 🔌 Multiple Transports — Stdio, HTTP, and SSE for any integration scenario
• 🔑 Dual API Support — Works with both Vertex AI (ADC) and Gemini API (API key)

Example Outputs

Prompt: "A photorealistic cat sitting in the rain on a cobblestone street, cinematic lighting"

🎬 Video Generation

Prompt: "A drone shot flying over a misty mountain valley at sunrise, cinematic" — Model: veo-3.1-generate-preview

🎵 Music Generation

Prompt: "An epic cinematic orchestral piece about a journey home. Starts with a solo piano intro, builds through sweeping strings." — Model: lyria-3-pro-preview

🗣️ Speech Synthesis

Prompt: "Say cheerfully: Have a wonderful day!" — Voice: Kore, Model: gemini-2.5-flash-preview-tts

🎭 Meme Generation

Prompt: "a cat sitting at a computer looking confused at code" — Top: "WHEN THE CODE WORKS", Bottom: "BUT YOU DONT KNOW WHY"

🎞️ GIF Generation

Prompt: "A cute robot waving hello, colorful cartoon style, looping animation"

📱 Short-Form Video

Prompt: "A timelapse of a flower blooming in a garden, vibrant colors, close-up macro shot" — Caption: "Nature is beautiful 🌸"

🎙️ Podcast Generation

Script: Host (Kore): "Welcome to AI Weekly!" → Guest (Puck): "Thanks for having me..." — Background: "soft ambient electronic"

📊 Presentation Generation

Slides: "ADK Rust MCP Toolkit" → "Dual API Support" → "Seven Servers" — Voice: Kore, Music: "soft ambient corporate"

📐 Diagram Generation

Prompt: "CI/CD pipeline: developer pushes code, triggers build, runs tests, if tests pass deploy to staging, then deploy to production" — Format: SVG, Type: flowchart

🎨 Artist Tools

Artist: "A peaceful countryside village at sunset with rolling hills and a winding river" — Style: oil_painting

Graphics Edit: "Add a hot air balloon floating in the sky" — Input: the oil painting above

📚 Education Tools

Whiteboard — Step-by-step math solutions and diagrams:

Prompt: "Solve step by step: 2x + 5 = 15" — Style: whiteboard, show_steps: true

Flashcards — Visual Q&A cards for any topic:

Topic: "Solar system planets" — Count: 3, Age group: 7-9

Quiz — Multiple-choice with illustrated questions:

Topic: "Dinosaurs" — Questions: 3, Difficulty: easy, Age group: 8-10

Story & Explainer — Narrated illustrated videos:

Story: "A brave little turtle who learns to swim" — Style: watercolor, Voice: Aoede, Moral: "Practice makes perfect"

Explainer: "How does the water cycle work?" — Style: diagram, Pace: slow, Age group: 8-10

Servers

Quick Start

Prerequisites

• Rust 1.94.1+ (2024 edition)
• Google Cloud project with Vertex AI enabled, or a Gemini API key
• gcloud CLI authenticated (for Vertex AI)
• FFmpeg (for avtool only)

Installation

# From crates.io
cargo install adk-rust-mcp-image adk-rust-mcp-video adk-rust-mcp-music \
              adk-rust-mcp-speech adk-rust-mcp-multimodal adk-rust-mcp-avtool

# Or build from source
git clone https://github.com/zavora-ai/adk-rust-mcp-toolkit
cd adk-rust-mcp-toolkit
cargo build --release

Configuration

# Option 1: Gemini API (recommended for getting started)
export GEMINI_API_KEY=your-api-key  # from https://aistudio.google.com/apikey

# Option 2: Vertex AI (for production/enterprise)
export PROJECT_ID=your-gcp-project
export LOCATION=us-central1
export GCS_BUCKET=your-bucket  # Required for video generation on Vertex

The servers auto-detect which API to use: if GEMINI_API_KEY is set, they use the Gemini API; otherwise they use Vertex AI with Application Default Credentials.

Run a Server

# Stdio (default) — for Claude Desktop, Kiro, local tools
adk-rust-mcp-image

# HTTP — for web apps, remote clients, ADK agents
adk-rust-mcp-image --transport http --port 8080

# SSE — for streaming applications
adk-rust-mcp-image --transport sse --port 8080

HTTP and SSE expose a strict, session-free MCP 2026-07-28 endpoint at /mcp. Each request must carry protocol/client metadata; initialization and Mcp-Session-Id are not accepted there. For a time-bounded migration, set ENABLE_LEGACY_MCP=1 to expose the old lifecycle separately at /mcp/legacy.

The video server pilots SEP-2663 Tasks for every Veo generation/extension call when the client advertises the extension. It returns a task handle immediately, supports polling and cooperative cancellation, suggests a one-second poll interval, and retains completed results for one hour. Legacy clients keep the synchronous call shape. The current rmcp task store is process-local, so task polling survives reconnects to that process but not a server restart or a hop to another replica. Tool discovery responses carry a one-hour public cache TTL.

Integration

Kiro

Add to .kiro/settings/mcp.json:

{
  "mcpServers": {
    "adk-image": {
      "command": "adk-rust-mcp-image",
      "args": ["--transport", "stdio"],
      "cwd": "/path/to/workspace",
      "env": {
        "PROJECT_ID": "your-project",
        "LOCATION": "us-central1"
      }
    }
  }
}

Note: The cwd field is required for file output with relative paths.

Claude Desktop

Add to ~/.config/claude/claude_desktop_config.json:

{
  "mcpServers": {
    "image": {
      "command": "adk-rust-mcp-image",
      "env": { "PROJECT_ID": "your-project" }
    }
  }
}

HTTP Client (Rust)

use adk_tool::McpHttpClientBuilder;

let toolset = McpHttpClientBuilder::new("http://localhost:8080/mcp")
    .timeout(Duration::from_secs(120))
    .connect()
    .await?;

{
  "mcpServers": {
    "adk-image": {
      "command": "adk-rust-mcp-image",
      "args": ["--transport", "stdio"],
      "cwd": "/workspace",
      "env": { "PROJECT_ID": "my-project", "LOCATION": "us-central1", "GCS_BUCKET": "my-bucket" }
    },
    "adk-video": {
      "command": "adk-rust-mcp-video",
      "args": ["--transport", "stdio"],
      "cwd": "/workspace",
      "env": { "PROJECT_ID": "my-project", "LOCATION": "us-central1", "GCS_BUCKET": "my-bucket" }
    },
    "adk-music": {
      "command": "adk-rust-mcp-music",
      "args": ["--transport", "stdio"],
      "cwd": "/workspace",
      "env": { "PROJECT_ID": "my-project", "LOCATION": "us-central1" }
    },
    "adk-speech": {
      "command": "adk-rust-mcp-speech",
      "args": ["--transport", "stdio"],
      "cwd": "/workspace",
      "env": { "PROJECT_ID": "my-project" }
    },
    "adk-multimodal": {
      "command": "adk-rust-mcp-multimodal",
      "args": ["--transport", "stdio"],
      "cwd": "/workspace",
      "env": { "PROJECT_ID": "my-project", "LOCATION": "us-central1" }
    },
    "adk-avtool": {
      "command": "adk-rust-mcp-avtool",
      "args": ["--transport", "stdio"],
      "cwd": "/workspace"
    }
  }
}

Architecture

┌─────────────────────────────────────────────────────────────┐
│                     MCP Servers                              │
│  image │ video │ music │ speech │ multimodal │ avtool       │
├─────────────────────────────────────────────────────────────┤
│                  adk-rust-mcp-common                         │
│         Config │ Auth │ GCS │ Transport │ Tracing           │
├─────────────────────────────────────────────────────────────┤
│                  Provider Abstraction                        │
├──────────────┬──────────────┬──────────────┬────────────────┤
│   Google     │    AWS       │   Azure      │    Local       │
│  Vertex AI   │  Bedrock     │  OpenAI      │   Ollama       │
│  Cloud TTS   │   Polly      │   TTS        │   Whisper      │
│   Gemini     │   Nova       │   GPT-4o     │                │
└──────────────┴──────────────┴──────────────┴────────────────┘

Currently Implemented: Google Cloud (Vertex AI, Cloud TTS, Gemini)

Planned: AWS Bedrock, Azure OpenAI, local/self-hosted models

Documentation

Testing

# Unit tests
cargo test --workspace

# Integration tests (requires GCP credentials)
cargo test --workspace --test integration_test

# Skip integration tests
SKIP_INTEGRATION_TESTS=1 cargo test --workspace

Contributing

See CONTRIBUTING.md for development setup and guidelines.

Security

See SECURITY.md for reporting vulnerabilities.

Contributors

License

Apache-2.0 — see LICENSE for details.

Built with ❤️ by Zavora AI

rmcp and MCP compatibility

This server is built with rmcp 3.1.2 and requires Rust 1.94.1 or newer. The rmcp 3 rollout retains legacy MCP initialization compatibility and targets MCP protocol revisions 2025-11-25 and 2026-07-28.