Skip to main content
LiteLLM supports two separate ways to use Parallel: its native Search provider or a connection to the Parallel MCP server through the LiteLLM MCP Gateway. The native Search provider does not require MCP.

Native LiteLLM Search provider

Install LiteLLM and set your Parallel API key:

Python SDK

Use search_provider="parallel_ai" to call Parallel Search through the LiteLLM Python SDK:
For more information, see LiteLLM’s Parallel Search provider documentation.

LiteLLM AI Gateway

Install the proxy dependencies and set a separate master key to authenticate requests to your local LiteLLM gateway:
Create a config.yaml file that explicitly registers parallel-search:
Start the gateway:
In another terminal, set the same LITELLM_MASTER_KEY and verify the tool is registered without making a Parallel Search request:
To run a real search, call the explicitly named parallel-search endpoint:

Parallel MCP through LiteLLM MCP Gateway

The LiteLLM MCP Gateway can also connect to the Parallel MCP server. This is separate from the native Search provider. Without an API key, the MCP tools the server exposes include web_search, web_fetch, and web_decisions. Add the server to your existing LiteLLM gateway configuration with HTTP transport:
The connection is anonymous by default. To enable optional Bearer authentication for higher rate limits, set PARALLEL_API_KEY and uncomment both authentication settings. With a key, the server also exposes the web_research tool. For more information, see LiteLLM’s MCP Gateway documentation and the Parallel MCP quickstart.