IDE and MCP client setup

Recommended setup

Use each tool's OpenAI-compatible provider setting first. Add the SeaLink base URL and your SeaLink API Key; model access, usage, and billing stay attached to the same account.

Generic configuration

export OPENAI_BASE_URL="https://api.sealink.io/v1"
export OPENAI_API_KEY="<your SeaLink API Key>"

If a client asks for the full chat endpoint, use https://api.sealink.io/v1/chat/completions. If it asks for a base URL, use https://api.sealink.io/v1.

Cherry Studio model setup

MCP adds SeaLink tools, but it does not add models to Cherry Studio's model selector. Configure SeaLink as a separate custom model provider.

1. Open Settings → Model Providers, select Add, and choose the OpenAI provider type.
2. Set the API address to https://api.sealink.io and enter your SeaLink API Key.
3. Select Get Model List, add the required models in Model Management, then enable the provider.

https://api.sealink.io/v1/models

If the provider connects but no model is selectable, confirm that the models were added in Model Management and that the provider switch is enabled.

MCP endpoint

MCP access is available to paid accounts. Connecting and listing tools are not charged. Successful billed tool and model calls use the same account balance, usage record, and billing journal as direct API calls.

CLI templates reference SEALINK_API_KEY without embedding it. For Cherry Studio and other clients that use a local placeholder, replace replace-with-your-sealink-api-key with the API Key. Do not put translated placeholder text in Authorization: HTTP header values must contain ASCII characters only.

Primary setup covers Cherry Studio, Claude Code, Cursor, Codex, WorkBuddy, and Windsurf. Check import and connection behavior in the exact client version used by the customer; a template is not proof of a successful connection.

{
  "mcpServers": {
    "sealink": {
      "type": "streamableHttp",
      "url": "https://api.sealink.io/api/v1/mcp",
      "headers": {
        "Authorization": "Bearer replace-with-your-sealink-api-key"
      }
    }
  }
}

{
  "mcpServers": {
    "sealink": {
      "type": "http",
      "url": "https://api.sealink.io/api/v1/mcp",
      "headers": {
        "Authorization": "Bearer ${SEALINK_API_KEY}"
      }
    }
  }
}

{
  "mcpServers": {
    "sealink": {
      "url": "https://api.sealink.io/api/v1/mcp",
      "headers": {
        "Authorization": "Bearer ${env:SEALINK_API_KEY}"
      }
    }
  }
}

[mcp_servers.sealink]
url = "https://api.sealink.io/api/v1/mcp"
bearer_token_env_var = "SEALINK_API_KEY"

{
  "mcpServers": {
    "sealink": {
      "type": "streamableHttp",
      "url": "https://api.sealink.io/api/v1/mcp",
      "headers": {
        "Authorization": "Bearer ${SEALINK_API_KEY}"
      }
    }
  }
}

{
  "mcpServers": {
    "sealink": {
      "serverUrl": "https://api.sealink.io/api/v1/mcp",
      "headers": {
        "Authorization": "Bearer ${env:SEALINK_API_KEY}"
      }
    }
  }
}

Image-to-video workflow

All six primary clients use the same ordered workflow. Only create_video_task requires per-call approval and charges after a successful billed creation; discovery, pricing, quoting, status, and recovery are not charged by the MCP wrapper.

1. connect
2. list_video_models
3. get_video_model
4. asset
5. quote_video
6. create_video_task · per-call approval
7. get_video_task
8. list_video_tasks

Built-in MCP tools

The tool list is filtered by the current API Key permissions, so some tools may be unavailable on entry plans.

• model_call · Calls models through /v1/chat/completions with the current API Key.
• list_models · Returns models callable by the current API Key.
• get_model_pricing · Returns current public model prices.
• get_balance_usage · Returns balance, today's usage, and recent usage.
• pii_scan · Scans and redacts PII for enterprise guardrail workflows.
• api_key_setup_guide · Generates MCP setup snippets for the six primary clients and compatible advanced clients.
• docs_search · Searches SeaLink docs and integration guides first.
• token_count · Counts tokens for text before a model call.
• web_search · Searches the web when enabled for the current plan.
• web_fetch · Fetches allowed URLs with SeaLink egress checks.
• code_exec · Runs Python or JavaScript in the sandboxed tool runner.
• image_generate · Generates images through the image API when enabled for the current plan.

Prompt templates are available through prompts/list and prompts/get. Resource subscriptions are intentionally unsupported on the stateless HTTP gateway; clients should re-read resources when they need fresh catalog data.