AI and MCP integration

Connect Dataset API to AI agents through its secured MCP server.

6 minDocs version 1.0

In this guide

1Connect the MCP server
2Grant least-privilege scopes
3Approve write operations

In this guide

1Connect the MCP server
2Grant least-privilege scopes
3Approve write operations

Connect the MCP server

The server is available at https://datasets.nordicdevhouse.com/api/mcp. Remote clients discover OAuth automatically and request only the scopes required by the selected tools.

JSON

{
  "type": "mcp",
  "server_label": "dataset-api-mcp",
  "server_url": "https://datasets.nordicdevhouse.com/api/mcp",
  "require_approval": {
    "always": {
      "read_only": false
    },
    "never": {
      "read_only": true
    }
  }
}

Client-specific setup

ChatGPT: In ChatGPT developer mode, create a custom MCP connection and paste the endpoint below. OAuth starts automatically.

URL

https://datasets.nordicdevhouse.com/api/mcp

Claude / Claude Code: Add the remote HTTP server in Claude or Claude Code, then complete OAuth with /mcp.

Shell

claude mcp add --transport http dataset-api-mcp https://datasets.nordicdevhouse.com/api/mcp

Cursor: Add this server to your project or user MCP configuration.

JSON

{
  "mcpServers": {
    "dataset-api-mcp": {
      "url": "https://datasets.nordicdevhouse.com/api/mcp"
    }
  }
}

VS Code: Add this server to .vscode/mcp.json or your user MCP configuration.

JSON

{
  "servers": {
    "dataset-api-mcp": {
      "type": "http",
      "url": "https://datasets.nordicdevhouse.com/api/mcp"
    }
  }
}

Codex: Add the remote Streamable HTTP server with the Codex CLI. Codex stores the connection in your MCP configuration.

Shell

codex mcp add dataset-api-mcp --url https://datasets.nordicdevhouse.com/api/mcp

OpenAI Responses API: Use the remote MCP tool in the Responses API and keep approval enabled for non-read operations.

JSON

{
  "type": "mcp",
  "server_label": "dataset-api-mcp",
  "server_url": "https://datasets.nordicdevhouse.com/api/mcp",
  "require_approval": {
    "always": {
      "read_only": false
    },
    "never": {
      "read_only": true
    }
  }
}

Generated tools

Tool	Mode	Scope	Approval
list_datasets	read	datasets:read	never
get_dataset	read	datasets:read	never
list_dataset_releases	read	datasets:read	never
get_dataset_release	read	datasets:read	never
query_dataset	read	query:read	never
download_dataset_release	read	datasets:read	never
list_dataset_builds	read	jobs:read	never
start_dataset_build	write	builds:write	required
get_dataset_build	read	jobs:read	never
cancel_dataset_build	destructive	builds:write	required
list_synthetic_datasets	read	datasets:read	never
create_synthetic_dataset	write	synthetic:write	required
get_synthetic_dataset	read	datasets:read	never
run_synthetic_dataset	write	synthetic:write	required
list_synthetic_dataset_releases	read	datasets:read	never
list_synthesis_runs	read	jobs:read	never
get_synthesis_run	read	jobs:read	never
get_dataset_api_health	read	datasets:read	never

MCP 2026 runtime

The server is stateless. Use server/discover, follow long-running operations through the Tasks extension, and answer input_required confirmations without changing the sealed requestState value.

Safety model

1
Grant only the API scopes the agent needs
2
Keep the API key server-side
3
Fulfil the signed confirmation round for write and destructive tools
4
Honor returned quota, retry, and circuit-breaker metadata
5
Preserve request and trace identifiers for auditing

Was this page helpful?