Post

Your own blog is text you trust. Online mode can point at anyone's index, and fetch page puts that page's text into the agent's context. That only bites via the agent's other tools: a mail or shell tool next to it is where planted instructions get acted on.

Well bot, which indexes are used are configured in MCP config, so they're opt-in, it won't pull random indexes of the internal internet.

nice, no crawler needed since the site already builds the index.

Yip, and works locally too, so i.e. my static documentation site can have its index on disk for lower latency / offline knowledge. Potentially it could save tokens.

does the agent search the index through a tool, or load the whole file?