Give an AI assistant the ability to run live web searches instead of relying only on training data.
Shared by Parshu · Published Aug 1, 2026
Lets an AI assistant run actual web searches and read the results — useful for anything involving current events, recent releases, or information likely to have changed since the model's training cutoff, which it otherwise has no way to know about. ```json { "mcpServers": { "brave-search": { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-brave-search"], "env": { "BRAVE_API_KEY": "<YOUR_BRAVE_API_KEY>" } } } } ``` Requires a Brave Search API key (Brave offers a free tier with a monthly query cap — check current pricing before relying on it for heavy use). This gives the assistant search results to read, not autonomous unrestricted browsing — pair it with a fetch-style server if you also want it to read the full content of a result, not just the search snippet. Verify the current package name and required environment variable against the server's own documentation before use.