Docs / Integrations

Integration guides

Connect Soundside to your preferred MCP client. All integrations use the same endpoint and auth model — the setup differs by client.

Endpoint

Common setup

MCP endpoint
https://mcp.soundside.ai/mcp
Discovery call
POST https://mcp.soundside.ai/mcp
Authorization: Bearer <token>
Content-Type: application/json

{"jsonrpc":"2.0","id":"1","method":"tools/list","params":{}}

Auth options: OAuth 2.1 (recommended for external clients) or API keys (for first-party integrations). See /docs/auth and /docs/api-keys.

Claude Code

Connect Soundside to Claude

Add Soundside as a custom connector in Claude. Open Settings → Connectors, click Add custom connector, and enter:

Name:     Soundside.ai
URL:      https://mcp.soundside.ai/mcp

Click Add to register the connector. You'll be redirected to authenticate with your Soundside account.

Add custom connector dialog showing Soundside.ai name and https://mcp.soundside.ai/mcp URL

Verify and configure permissions

After authenticating, return to Settings → Connectors. You should see soundside.ai listed. Click into it to manage tool permissions:

  • • Read-only tools (3) — analysis, library browsing, sharing.
  • • Write tools (9) — generation, editing, library management.

We recommend Always allow for read-only, Ask first for write/delete until you're comfortable.

Soundside.ai connector detail page showing tool permission controls

Try it

Start a new conversation in Claude and try:

Create a cinematic image of a sunset over the ocean with dramatic cloud formations.

Troubleshooting

  • • Connector not appearing: Verify the exact URL https://mcp.soundside.ai/mcp and complete the OAuth flow.
  • • Tool denied: Check permission settings — a tool set to "Never allow" won't execute.
  • • Session expired: Reconnect the connector in Settings → Connectors.
  • • Credit errors: Check your balance on the billing page.

Anthropic

Anthropic connector guide

Use this setup for Anthropic-compatible environments that support external MCP tools and OAuth-based access.

Bearer-authenticated tools/list
{
  "jsonrpc": "2.0",
  "id": "1",
  "method": "tools/list",
  "params": {}
}

Error handling

Handle policy denials (forbidden_client) distinctly from quota denials; policy failures require trust configuration, not retry.

Safe rollout defaults

  • • Keep tool lists explicit and minimal at start.
  • • Monitor first-call failure rates before widening surfaces.
  • • Use project-scoped library operations for least privilege.

OpenAI

OpenAI remote MCP guide

Apply this pattern when your agent runtime supports remote MCP tool connectivity and you want Soundside tools as external capabilities.

Initial discovery call
POST https://mcp.soundside.ai/mcp
Authorization: Bearer <token>
Content-Type: application/json

{"jsonrpc":"2.0","id":"1","method":"tools/list","params":{}}

Error handling

Map retryable and quota headers directly into your retry policy. Avoid blind exponential retries on non-retryable denials.

Recommended launch profile

  • • Start with read-only tools in early rollout.
  • • Add generation tools once prompt quality and budget controls stabilize.
  • • Reserve advanced workflows for explicit operator-approved automations.

OpenClaw

OpenClaw agentic media guide

Connect OpenClaw to Soundside for generation, editing, analysis, and library management with durable managed assets.

Provider coverage through one endpoint
create_video: vertex (Veo 3.1), runway, minimax, luma, grok
create_image: vertex, grok, runway, minimax, luma
create_audio: minimax, vertex
create_music: minimax
create_text: vertex (Gemini), grok, minimax
edit_video: soundside.ai (21 editing actions)
analyze_media: soundside.ai + Gemini vision QA

Durable resource pattern

Resource-centric chaining
1) Generate media -> receive resource_id
2) Store/organize via lib_manage
3) Reuse resource_id in downstream tools
4) Query status with lib_list for resumed sessions
5) Download only for final delivery or external handoff

This reduces local storage pressure and keeps workflow state durable across sessions.

Execution guardrails

  • • Follow tools/list schemas exactly instead of hardcoding argument assumptions.
  • • Respect Retry-After and retryable instead of blind retries.
  • • Track per-tool success and credit burn to tune routing and approval policies.