Description
Would an opt-in Parallel Search MCP integration fit the template? It would give agents free web search and page fetching without a Parallel account or API key. Search uses Fast mode, suited to finding current information beyond the synced knowledge base.
I propose adding optional MCP-backed tools in @savoir/agent, with an explicit selection in createAgent and createSourceAgent. The existing OpenAI web-search tool would remain the default. The new tools would use the HTTP MCP endpoint, preserve the loading/result generator format used by the chat UI, and support cancellation and bounded timeouts. Validation would exercise tool dispatch through the agent loop, returned excerpts and source URLs, and the anonymous path.
Additional context
This would supplement the existing sandbox tools, without changing knowledge-source syncing. The closed MCP questions #27 and #28 concern exposing TODO-app tools to external clients; this proposal concerns consuming search and fetch tools inside the agent.
I work at Parallel.
Activity
0No activity yet.