MCP Providers | NVIDIA NeMo Data Designer

MCP providers are external services that host and serve tools via the Model Context Protocol (MCP). Data Designer uses provider configurations to establish connections to these services.

Overview

An MCP provider defines how Data Designer connects to a tool server. Data Designer supports two provider types:

Provider Class	Connection Method	Use Case
`MCPProvider`	SSE or Streamable HTTP	Connect to a pre-existing MCP server
`LocalStdioMCPProvider`	Subprocess via stdin/stdout	Launch an MCP server as a subprocess

When you create a ToolConfig, you reference providers by name, and Data Designer uses those provider settings to communicate with the appropriate MCP servers.

MCPProvider (Remote)

Use MCPProvider to connect to a pre-existing MCP server. Both SSE (Server-Sent Events) and Streamable HTTP transports are supported:

1 import data_designer.config as dd
2 from data_designer.interface import DataDesigner
3 
4 # SSE transport (default)
5 sse_provider = dd.MCPProvider(
6     name="remote-mcp",
7     endpoint="http://localhost:8080/sse",
8     api_key="MCP_API_KEY",  # Environment variable name
9 )
10 
11 # Streamable HTTP transport
12 http_provider = dd.MCPProvider(
13     name="remote-tools",
14     endpoint="https://mcp.example.com/mcp",
15     api_key="MCP_API_KEY",
16     provider_type="streamable_http",
17 )
18 
19 data_designer = DataDesigner(mcp_providers=[sse_provider, http_provider])

MCPProvider Fields

Field	Type	Required	Description
`name`	`str`	Yes	Unique identifier for the provider
`endpoint`	`str`	Yes	Endpoint URL for the remote MCP server
`api_key`	`str`	No	API key or environment variable name
`provider_type`	`str`	No	Transport type: `"sse"` (default) or `"streamable_http"`

LocalStdioMCPProvider (Subprocess)

Use LocalStdioMCPProvider to launch an MCP server as a subprocess:

1 import data_designer.config as dd
2 from data_designer.interface import DataDesigner
3 
4 mcp_provider = dd.LocalStdioMCPProvider(
5     name="demo-mcp",
6     command="python",
7     args=["-m", "my_mcp_server_module"],
8     env={"MY_SERVICE_TOKEN": "..."},
9 )
10 
11 data_designer = DataDesigner(mcp_providers=[mcp_provider])

LocalStdioMCPProvider Fields

Field	Type	Required	Description
`name`	`str`	Yes	Unique identifier for the provider
`command`	`str`	Yes	Executable to run (e.g., `"python"`, `"node"`)
`args`	`list[str]`	No	Command-line arguments
`env`	`dict[str, str]`	No	Environment variables for the subprocess
`provider_type`	`str`	No	Always `"stdio"` (set automatically)

API Key Configuration

The api_key field can be specified in two ways:

Environment variable name (recommended): Set api_key to the name of an environment variable (e.g., "MCP_API_KEY"). Data Designer will resolve it at runtime.
Plain-text value: Set api_key to the actual API key string. This is less secure and not recommended for production.

1 # Method 1: Environment variable (recommended)
2 provider = dd.MCPProvider(
3     name="secure-mcp",
4     endpoint="https://mcp.example.com/sse",
5     api_key="MCP_API_KEY",  # Will be resolved from environment
6 )
7 
8 # Method 2: Direct value (not recommended)
9 provider = dd.MCPProvider(
10     name="secure-mcp",
11     endpoint="https://mcp.example.com/sse",
12     api_key="actual-api-key-value",
13 )

YAML Configuration

Both provider types use a provider_type discriminator field in YAML configurations. When writing YAML configs manually (e.g., in ~/.data-designer/mcp_providers.yaml), include the discriminator:

1 providers:
2   # Remote SSE provider
3   - name: doc-search
4     provider_type: sse
5     endpoint: http://localhost:8080/sse
6     api_key: ${MCP_API_KEY}
7 
8   # Remote Streamable HTTP provider
9   - name: remote-tools
10     provider_type: streamable_http
11     endpoint: https://mcp.example.com/mcp
12     api_key: ${MCP_API_KEY}
13 
14   # Local stdio provider
15   - name: local-tools
16     provider_type: stdio
17     command: python
18     args:
19       - -m
20       - my_mcp_server
21     env:
22       DEBUG: "true"

Using Multiple Providers

You can configure multiple MCP providers and use them together in a single ToolConfig:

1 import data_designer.config as dd
2 from data_designer.interface import DataDesigner
3 
4 providers = [
5     dd.MCPProvider(
6         name="doc-search-mcp",
7         endpoint="http://localhost:8080/sse",
8     ),
9     dd.LocalStdioMCPProvider(
10         name="calculator-mcp",
11         command="python",
12         args=["-m", "calculator_mcp"],
13     ),
14 ]
15 
16 data_designer = DataDesigner(mcp_providers=providers)