PentestGPT's interactive google command currently returns an under-development message. This makes that search path useful: choose search in the main or local task menu, enter a query, and get linked results from Parallel's free Search MCP. The old google command still works as an alias and clearly says it uses Parallel. No key or provider setup is needed.
The anonymous endpoint uses Fast search at lower rate limits, per the Search MCP docs. Parallel Fast also appears in independent DeepSearchQA results and Time per Task charts. Those charts describe the Search API, not PentestGPT's runtime or this MCP integration.
The change uses the project's existing Python MCP dependency, sends the user's query to web_search, and shows up to five results with source URLs and bounded excerpts. Requests identify the project as PentestGPT/1.0.0 mcp-python in the User-Agent so Parallel can measure aggregate free MCP usage. The CLI help tells users that their query goes to Parallel. This doesn't change the autonomous agent or trigger search without someone choosing the command.
I ran the root test suite (136 passed), Ruff check and formatting, and a hosted search through the interactive command handler. A local MCP transport test checked the outgoing User-Agent and confirmed the anonymous requests have no Authorization header. I haven't verified the aggregate usage in Parallel analytics.
Btw - I'm a devrel engineer at Parallel