soundside.ai

official

MCP-native AI media generation with x402 pay-per-call. Image, video, audio, and music from 6 providers — composable via resource IDs. USDC on Base.

GitHubTry this MCPSponsored

What can you do with Soundside Ai MCP?

Generate media assets — Ask for images, video, audio, music, or text via create_image, create_video, create_audio, create_music, or create_text, with provider selection.
Edit and compose media — Trim, concat, color grade, or burn subtitles with edit_video; mix audio with edit_audio; build split-screen composites via compose_media.
Analyze media content — Extract technical metadata, transcribe audio, detect segments, or export an EDL using analyze_media.
Train and manage LoRA adapters — Fine-tune an adapter from library media with train_adapter, then list, deploy, or delete adapters via list_adapters and manage_adapter.
Organize and share projects — Browse projects, collections, and brand kits with lib_list; create or update resources via lib_manage; share projects by email using lib_share.

Documentation

Soundside — Developer Documentation

AI Media Production Platform for Agents

Soundside exposes 19 MCP tools for generating, editing, composing, extracting, and analyzing media — images, video, audio, music, text, and business artifacts — plus LoRA adapter fine-tuning and server-side video composition. Connect any MCP client. OAuth/API-key credits cover all authenticated tools; the eligible subset also supports x402 USDC on Base without an account. Compose is authenticated-credit only.

Currency 2026-08 (2026-08-23)

Removed: Luma (entirely) and Runway image/video generation. Runway is now audio-only — TTS and sound effects via create_audio.

Added: Lyria 3 music generation (create_music), Grok TTS (create_audio), Grok per-second × resolution video pricing, Alibaba Wan 2.7 video models (international default), MiniMax H3 video adapter.

Provider and pricing data below matches the live x402 catalog: GET https://mcp.soundside.ai/api/x402/status.

Quick Start

# MCP endpoint
https://mcp.soundside.ai/mcp

# Auth: API key or x402 crypto payment
Authorization: Bearer <your-api-key>

POST https://mcp.soundside.ai/mcp
{"jsonrpc":"2.0","id":"1","method":"tools/list","params":{}}

Tools (19)

Generation

Tool	What It Does	Providers
`create_image`	Text-to-image, character references	Alibaba (Wan), Creative Freedom (authenticated only), Grok, MiniMax, Vertex AI
`create_video`	Text-to-video, image-to-video, video extension	Alibaba (Wan 2.7), Creative Freedom (authenticated only), Grok, MiniMax (Hailuo/H3), Vertex AI (Veo 3.1)
`create_audio`	TTS, sound effects, deprecated transcription shim, voice cloning/design	Creative Freedom (authenticated only), Grok, MiniMax, Runway (audio-only), Vertex AI
`create_music`	Music from lyrics and style prompts	Lyria 3; Creative Freedom (authenticated credits only)
`create_text`	LLM chat completions, structured output	Grok, MiniMax, Qwen, Vertex AI (Gemini)
`create_artifact`	Charts, presentations, documents, diagrams; bundle mode for multi-artifact packages	plotly, pptx, docx, weasyprint, mermaid, gamma

Composition

Tool	What It Does
`compose_video`	Server-side pipeline: enrich plan, generate assets in parallel, assemble with transitions, audio ducking, and overlays

Editing

Tool	What It Does
`edit_video`	Core video transforms: trim, concat, crossfade, speed, loop, color grade, burn subtitles, custom FFmpeg
`edit_audio`	Mix, replace, or pad audio on existing media
`compose_media`	Add text, overlay media, or build split-screen composites
`apply_effect`	Ken Burns, speed ramp, film grain, vignette
`extract_media`	Extract frames, frame sets, or audio tracks

Analysis

Tool	What It Does	Providers
`analyze_media`	Technical metadata, vision QA, transcription, segment detection, EDL export	Anthropic, Grok, OpenAI, Qwen, Vertex (+ soundside.ai ffprobe)

Adapters (LoRA)

Tool	What It Does	Backends
`train_adapter`	Train a LoRA adapter from library media	DashScope (Wan), Modal (Hunyuan/LTX)
`list_adapters`	List your LoRA adapters
`manage_adapter`	Inspect, deploy, undeploy, delete, or select checkpoint

Library Management

Tool	What It Does
`lib_list`	Browse projects, collections, resources, lineage, brand kits; query credit balance
`lib_manage`	CRUD for projects, collections, resources, brand kits
`lib_share`	Share projects with other users by email

Pricing

Soundside uses a credit system: one credit = $0.01 USD.

AI generation uses published metered rates based on provider cost with an approximately 10% platform margin unless a tool-specific flat fee is listed.
Platform tools (editing engine, library) are fixed-price: $0.01/call; vision QA is $0.03.
Compose adds a five-credit success-only orchestration fee and separately itemizes child calls. It requires OAuth/API-key credits and is not available through x402.
Every paid call receives a pre-execution estimate. The estimate is a ceiling — the actual charge is never more than the quote — and each paid tool call is settled exactly once.

Live pricing is always available at:

GET https://mcp.soundside.ai/api/x402/status

This returns machine-readable per-tool, per-provider USDC prices for the x402 lane only. It does not publish the free tool or authenticated-credit-only Compose. Prices are DB-driven and may change — always check the endpoint rather than hardcoding. For variable-priced tools the published amount is a ceiling quote (worst case), not the typical settled price — rows carry a price_note where this matters.

x402: Pay-Per-Call with Crypto

No API key needed. Pay with USDC on Base (L2) per tool call via EIP-3009 transferWithAuthorization (off-chain signing, facilitator pays gas).

Network: eip155:8453 (Base mainnet)
Token: USDC
Facilitator: Coinbase CDP

See x402 Guide for full setup.

Guides

Getting Started — First MCP connection in 5 minutes
x402 Pay-Per-Call — Crypto payments, no account needed
Tool Reference — Detailed docs for all 19 tools

Examples

Python — API Key — Connect and generate with httpx
Python — x402 — Pay-per-call with USDC
TypeScript — API Key — Node.js MCP client
OpenClaw Skill — One-line config for OpenClaw agents

Local validation (pre-push hook)

There is no GitHub-hosted CI; validation runs locally before every push. Enable once per clone:

git config core.hooksPath .githooks

Requirements: brew install gitleaks and uv. The hook runs three gates on every push: (1) a gitleaks secret scan of the outgoing commit range, (2) python3 scripts/validate_public_contract.py, and (3) the pytest suite (uv run --with pytest python -m pytest tests/ -q). The contract validator compares this portal against the vendored MCP contract and its source of truth in the sibling repos, so pushing requires the three-repo workspace checkout (../ssd-mcp and ../soundside-ai next to this repo); a standalone clone cannot pass the gate.