When to use Parallel MCPs?
As can be seen in the following table, our MCPs (see installation instructions here) can be useful for quick experimentation with deep research and task groups, or for daily use. It can be a great way to get to know Parallel’s different APIs by exploring their capabilities.| Use Case | What |
|---|---|
| Agentic applications where low-latency search is a tool call | Search MCP |
| Daily use for everyday deep-research tasks in chat-based clients | Task MCP |
| Enriching a dataset (eg. a CSV) with web data via chat-based clients | Task MCP |
| Running benchmarks on Parallel processors across a series of queries | Task MCP |
| Building high-scale production apps that integrate with Parallel APIs | Search & Task APIs |
- Currently, since data needs to go through the context window, the MCP is not able to perform large-scale task groups.
- As the MCP / Connector landscape is still evolving, there are a limited number of well-working data sources and destinations.
- Currently, it is not possible to complete a series of task groups and deep research from a single prompt in one go, and you also cannot be notified. To perform multiple tasks or batches in a workflow, you need to reply each time to verify the task is complete and initiate the next step. We are working on improving this.
- While our Search MCP is designed to work well with smaller models as well (such as GPT OSS 20B), our Task MCP is recommended for use with larger models only (such as GPT-4o or Claude Sonnet 3.5).