One prompt.Whole broadcast.

Install the MCP server.

Add MangoPods as a remote MCP server in the agent environment. Developers can use an API key for personal automation or OAuth for user-delegated access.

{
  "mcpServers": {
    "mangopods": {
      "url": "https://api.mangopods.com/mcp",
      "headers": {
        "Authorization": "Bearer ${MANGOPODS_API_KEY}",
        "MangoPods-Workspace": "${MANGOPODS_WORKSPACE_ID}"
      }
    }
  }
}

First test call.

{
  "tool": "workspace.get_profile",
  "arguments": {}
}

Auth and scopes.

MangoPods apps request the smallest scope set that can complete the job. Read scopes let agents inspect existing projects and assets. Write scopes can create, render, publish, spend credits, or manage team operations.

Credit lifecycle.

Every meaningful job should start with an estimate. Agents can show the user expected spend, reserve a cap, execute the work, then settle or refund unused credits.

1. Call credits.estimate with the tool family, input assets, quality tier, duration, language count, render target, and publishing destinations.
2. Call credits.reserve with a maximum cap, user confirmation, and idempotency key.
3. Pass credit_hold_id into the creative tool call.
4. Watch the job until complete, failed, cancelled, or awaiting approval.
5. Call credits.settle, or let MangoPods auto-settle once the job is complete.

{
  "tool": "credits.estimate",
  "arguments": {
    "operation": "video.generate_clips",
    "quality": "pro",
    "source_asset_ids": ["asset_raw_interview_01"],
    "outputs": [
      {"type": "vertical_clip", "count": 12, "duration_seconds": 45},
      {"type": "captions", "languages": ["en"]},
      {"type": "thumbnail", "count": 12}
    ]
  }
}

Assets and media plane.

MCP should not haul giant video files through the conversation. MangoPods uses an asset model: agents create upload sessions, import media from URLs, attach metadata, and receive signed URLs for download or preview.

Jobs and states.

Creative operations are asynchronous by default. Short operations can return inline. Expensive work returns a job that can be watched, cancelled, retried, or resumed.

{
  "tool": "job.get",
  "arguments": {
    "job_id": "job_clipfarm_8uJ2",
    "include_logs": true,
    "include_outputs": true
  }
}

Resources.

Agents can read MCP resources before taking action. This keeps them from asking the user for information MangoPods already knows.

Webhooks.

Use webhooks when your app needs to react outside the current agent session. Each event is signed and retryable.

Error model.

Errors are designed for agents. Every failure includes a stable code, a human-readable explanation, retry guidance, and a suggested next tool.

{
  "error": {
    "code": "credit_limit_exceeded",
    "message": "This render needs 184 credits, but the approved cap is 120.",
    "retryable": true,
    "suggested_next_tool": "credits.estimate",
    "safe_user_message": "The requested quality tier costs more than the approved budget. Try fewer clips, lower quality, or approve a higher cap."
  }
}

Security and reliability.

• Use OAuth for user-facing products and API keys only for owned workspace automation.
• Send an idempotency key for every spend, render, publish, booking, and import operation.
• Prefer signed asset URLs with short lifetimes instead of sending raw media through chat history.
• Use dry runs for destructive or expensive operations.
• Request user confirmation for public publishing, paid boosts, high credit spend, and outreach campaigns.
• Store webhook signing secrets outside the client and rotate them when a partner app is removed.