Skip to main content

MCP Web Search

· 3 min read
Basile Chevalier
Basile Chevalier
Project manager

🚀 Web Search is coming to Clovis.

We are launching our Web Search MCP, now available both in the chatbot and through the API. The goal: turn your model into a truly real-world connected assistant, capable of retrieving information wherever it is.

Thanks to the MCP integration, the model can now:

  • Automatically launch web searches
  • Analyze and summarize recent content
  • Provide richer, more reliable, and up-to-date answers

This improvement enables dynamic, contextualized interactions connected to current events.

🔌 Simple integration, endless possibilities

In just a few lines, you can allow your agents to access the web and leverage its full potential.

Agent

MCP connection

Web Search is exposed as an MCP server and connected directly to your application, agent, or orchestrator.

The model can then:

  • decide when to perform a web search,
  • call the MCP,
  • and use the results in its response.

👉 This approach provides maximum control and makes it possible to build:

  • autonomous agents,
  • advanced workflows,
  • multi-tool orchestrations,
  • or assistants connected to business systems.

It integrates easily into existing agentic architectures.

Link to the documentation

Model

Integrated into the model

Web Search can also be integrated directly into certain Clovis models to greatly simplify the integration experience.

With this approach, the model is directly able to:

  • detect when a web search is relevant,
  • automatically perform the search,
  • and use the results in its generation.

👉 Everything happens natively within the model’s reasoning, without any complex orchestration on the client side.

There is no need to manage fallback logic or specific routing.

Available on: ClovisLLM/gpt-oss-120b

Link to the documentation

This is only the beginning.

Web Search will continue to evolve with increasingly advanced capabilities: DeepSearch, multi-source analysis, intelligent browsing, and more.

Update made by
Basile ChevalierBasile Chevalier