Claude Research Mode
When you turn on advanced research, Claude hands your request to a dedicated research agent that works across the web and Google Drive. That agent runs in its own environment and does not have access to your connected MCP servers — including Basic Memory, Gmail, Google Calendar, and any other connectors. This is a limitation of research mode itself, not of any individual connector: every MCP connector is unavailable during a research run, in the same way.
What this means in practice:
- During a research run, Claude can't read from or write to Basic Memory (or call any other connector's tools). It will research using web and Drive sources only.
- After a research run finishes, control returns to the normal conversation — which does have your connectors. At that point Claude can save the results to Basic Memory, draft an email, add a calendar event, and so on.
So research and your connectors aren't combined in a single step, but they chain cleanly: research first, then act on the results with your connectors second.
Recommended flow: research → save to Basic Memory
Turn advanced research on
Toggle advanced research in the message composer, then run your query — for example, "survey how other agent-memory tools expose graph APIs."
Wait for the report
Let the research run complete. The report comes back inline in the conversation.
Turn advanced research off
Use the same toggle to drop back into normal conversation. This is the important step — leaving research on keeps connectors out of reach.
Ask Claude to save it
Now you're back in the normal tool path. Ask Claude to write up the results — "Save that as a note in my research project" — and it can call Basic Memory's write_note directly.
The same pattern works in reverse for context you already have: if you need Claude to use your Basic Memory notes as input, keep research off and ask normally — Claude can read your notes and reason over them directly. Use research only when you specifically want fresh web/Drive sources.
Why this happens
Research mode runs a separate agent with its own toolset (web search, Google Drive). MCP connectors live on the main Claude conversation, not inside the research agent's environment. The research agent doesn't know your connectors exist, so it can't call them — and there's no way to "pass through" a connector during a research run.
This is the same for every MCP connector, not specific to Basic Memory. If you have a workflow that wants both web research and your private knowledge base in a single step, today the answer is to run them sequentially: research first, then a follow-up turn that uses your connectors on the results.
Related
Hermes Plugin
Give Hermes Agent persistent Basic Memory: search-before-answer recall, per-turn capture, end-of-session summaries, 10 agent tools, slash commands, and a bundled skill.
Claude Code
Give Claude Code persistent memory with Basic Memory. Decisions, architecture, and project context carry over across every coding session.