Summary
One call for the fact you are missing, another for the write-up you were dreading.
The two tools are deliberately different jobs: deep_search answers a question that outran the model's knowledge cutoff, while comprehensive_research goes away and assembles an investigation across sources. Choosing between them is the whole interface, and modelClass is the dial on the expensive one.
What it is
A two-tool search server: a direct multi-provider web search, and an agent-driven research pass over a topic
What you get
- Real-time web search across multiple providers with the provider selectable per call, a result cap, and an optional direct answer alongside the results (deep_search).
- An agent pass that searches multiple sources, analyses what it finds and compiles a report, with modelClass choosing how much model to spend on it (comprehensive_research).
Requirements
ENV_FILE — the path to the environment file the server loads its configuration from.
Setup effort
One command plus a key — claude mcp add deep-search -s user -e ENV_FILE=$HOME/.llm.env -- npx -y @just-every/mcp-deep-search, then supply credentials