Summary
Finding a source and reading it are one hop apart, so a claim gets checked instead of repeated.
The two tools close the loop between a query and the actual page, which means an answer can cite the text it came from rather than a search snippet. num is the only lever on how much of the context window a research pass costs, and the scope of what is searchable is set once in the programmable search engine rather than per query.
What it is
Two tools that cover a research loop: search runs a web query through your own Google programmable search engine and returns results, and read_webpage fetches a URL and extracts its text content.
What you get
- A web search returning results, with num controlling how many come back
- The text content of any page, extracted from a URL you pass
- A search engine whose scope is yours: results reflect however the programmable search engine in GOOGLE_SEARCH_ENGINE_ID is configured
Requirements
A Google API key in GOOGLE_API_KEY and a programmable search engine ID in GOOGLE_SEARCH_ENGINE_ID, plus Node on your PATH.
Setup effort
One command plus a key — npx -y @adenot/mcp-google-search, then supply credentials