@agentutility/mcp-synthforge

Generative media for agents that ship products.

Image, video, music, voice generation across three price tiers. One API surface, USDC-settled, no SaaS account.

Pricing: pay-per-call in USDC on Base. No subscriptions, no API keys. See per-tool prices below.

Install — Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "agentutility-synthforge": {
      "command": "npx",
      "args": ["-y", "@agentutility/mcp-synthforge"],
      "env": { "X402_PRIVATE_KEY": "0xYOUR_PRIVATE_KEY_HEX" }
    }
  }
}

Restart Claude Desktop. 14 tools appear in the tool palette.

Install — Cursor

Add to .cursor/mcp.json:

{
  "mcpServers": {
    "agentutility-synthforge": {
      "command": "npx",
      "args": ["-y", "@agentutility/mcp-synthforge"],
      "env": { "X402_PRIVATE_KEY": "0x..." }
    }
  }
}

Funding

Send any amount of USDC on Base mainnet to the address derived from your X402_PRIVATE_KEY. The MCP server uses it to pay for tool calls automatically.

USDC on Base contract: 0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913

Tools (14)

| Tool | Description | |---|---| | image-edit | (0.02 USDC/call) Image edit / instruction-based image edit / text-driven photo edit / nano-banana image editor / GPT-image-2 edit. Mask-free instruction-driven image editing — describe the change in text and the model applies it to the whole image. Default model: nano-banana-pro. Returns a permanent fal-hosted PNG URL. | | image-expand | (0.15 USDC/call) AI image outpainting / image expansion. Bria Expand model. Generates realistic content beyond original borders. Set canvas size + original placement. Commercial-license model. | | image-generate | (0.01 USDC/call) Image generate (fast/cheap) / text-to-image / AI art. SFW, sub-5s turnaround. Four tiers backed by curated $0.01 Venice models: 'fast' (z-image-turbo, default), 'creative' (chroma), 'anime' (wai-Illustrious), 'sd35' (venice-sd35). For top-tier quality use image-generate-pro ($0.10, Flux 2 Pro / Recraft / Seedream / Qwen Image 2 Pro) or image-generate-ultra ($0.30, Google nano-banana-pro / OpenAI gpt-image-2). Returns a permanent fal-hosted PNG URL. | | image-generate-pro | (0.10 USDC/call) Image generate (pro) / premium text-to-image / Flux 2 Pro / Recraft / Seedream / Qwen Image 2 Pro / xAI Grok Imagine. Premium multi-model lineup for photoreal, design/illustration, text-in-image, and stylized art. Tiers: 'balanced' (flux-2-pro, default), 'max' (flux-2-max), 'text' (qwen-image-2-pro — best at rendering text in images), 'recraft' (recraft-v4), 'seedream' (seedream-v4), 'grok' (grok-imagine-image), 'art' (imagineart-1.5-pro), 'hunyuan' (hunyuan-image-v3). For flagship Google/OpenAI models use image-generate-ultra. Returns a permanent fal-hosted PNG URL. | | image-generate-ultra | (0.30 USDC/call) Image generate (ultra) / flagship text-to-image / Google nano-banana-pro (Gemini Image 3) / OpenAI gpt-image-2 / Recraft V4 Pro / xAI Grok SOTA. Top proprietary models for the highest quality output. Tiers: 'nano-banana' (nano-banana-pro, default — Google Gemini Image 3), 'nano-banana-2' (cheaper Google variant), 'gpt' (gpt-image-2 — OpenAI flagship), 'gpt-1-5' (gpt-image-1-5), 'recraft-pro' (recraft-v4-pro), 'grok-sota' (grok-imagine-image-quality). Output is capped to 1024x1024 to keep wholesale within retail; for higher resolution, chain image-upscale. Returns a permanent fal-hosted PNG URL. | | image-inpaint | (0.02 USDC/call) Image inpainting / mask-based image edit / fill in masked region / object replacement / face swap (mask-driven) / generative fill. Replaces the masked region of an image with content matching a text prompt. White pixels in the mask = region to inpaint. Default model: gpt-image-2. Returns a permanent fal-hosted PNG URL. | | image-to-video | (0.20 USDC/call) Image-to-video / animate still image / Seedance image-to-video / motion-from-photo / camera-movement on photo. Animates a still image into video via Venice's seedance-2-0-fast-image-to-video. Optional prompt steers the motion (camera moves, subject motion). Same async-vs-sync handling as text-to-video. | | music-generate | (0.05 USDC/call) Music generation / text-to-music / AI music / generative song / instrumental and vocal music. Text-to-music via Venice with the minimax-music-v26 model. Optional lyrics input. Duration 5-120 seconds. Returns a permanent fal-hosted audio URL (or a Venice-hosted URL when Venice already provides one). | | remove-bg | (0.08 USDC/call) AI background remover / background eraser / cutout tool. Returns transparent PNG. Optional crop_to_bbox. fal.ai imageutils/rembg. | | seedance-video | (0.20 USDC/call) Seedance 2.0 / Seedance 2.0 video generation / Seedance video AI / generative AI video / text-to-video AI / cinematic AI clips on AI Gateway. Powered by Venice's seedance-2-0-fast-text-to-video model. Duration / aspect-ratio / resolution configurable. Same backend as text-to-video under a model-named slug for direct discovery by agents searching for 'Seedance'. | | sound-effect-generate | (0.01 USDC/call) Sound effect generation / text-to-SFX / Foley generator / ElevenLabs sound effects / ambient audio synth. Text-to-SFX via Venice with elevenlabs-sound-effects-v2. Duration 0.5-22 seconds. Returns a permanent fal-hosted audio URL (or Venice-hosted when applicable). | | text-to-speech | (0.05 USDC/call) Text to speech / TTS / voice generator. Venice TTS (Kokoro / xAI / ElevenLabs / Orpheus / MiniMax / Gemini). 30+ voices, 6 audio formats. Returns hosted MP3 URL. | | text-to-video | (0.20 USDC/call) Text-to-video / AI video / Seedance / generative video / cinematic clip from prompt. Text-to-video via Venice's seedance-2-0-fast-text-to-video. Duration / aspect-ratio / resolution configurable. The synchronous path has a 22s budget; if Venice can't return inline within that window, the response surfaces a job_id + poll_url for the caller to resolve later. | | voice | (0.05 USDC/call) Text-to-speech / TTS / voice synthesis. Venice TTS (Kokoro/xAI/ElevenLabs/Orpheus/MiniMax). 30+ voices, MP3/WAV/OPUS/AAC/FLAC. |

How it works

1. Agent calls a tool (e.g. image-edit).
2. MCP server POSTs to https://x402.agentutility.ai/image-edit.
3. The endpoint responds HTTP 402 with payment instructions.
4. The MCP server signs an EIP-3009 USDC transfer authorization with X402_PRIVATE_KEY and retries.
5. CDP facilitator settles on Base.
6. The endpoint returns the actual response.

The agent never sees the payment flow — it just gets the result.

Links

• Cluster overview: https://agentutility.ai/synthforge/
• All MCP packages: https://mcp.agentutility.ai/
• Source: https://github.com/rooz21/x402/tree/main/packages/mcp-synthforge

Version: 0.1.8 · License: MIT