What does MCP do for AI memory?
A plain-language guide to MCP clients, servers, memory tools, and the difference between connecting an agent and capturing its work.
Understand AI memory 2 min read
How shared memory between AI assistants works
Shared memory gives more than one connected assistant access to a common source of context. It does not merge their accounts, model states, or private conversations. Each assistant still needs an authorized connection and must retrieve the information.
A shared source, separate readers
Two assistants can read the same project note and produce different answers. They may ask different search questions, retrieve different passages, or interpret the same instruction differently. Shared access reduces missing background; it does not make their reasoning identical.
The connection also has two sides. A source has to reach the memory store. Then a destination needs permission and working tools to read it. Testing only sign-in leaves most of that path untested.
Check the path with a harmless fact
Use a made-up project and a distinctive detail that is safe to share. Wait for capture and sync, then ask a second connected assistant to find it. Ask for the source as well as the detail.
If the assistant cannot find it, record which step failed. Avoid testing with private client material simply because it is easy to recognize. A short synthetic note works just as well.
- Confirm the source conversation was included.
- Confirm the destination is using the intended Dray account.
- Ask the destination to consult its memory tool explicitly.
TAKE THIS WITH YOU
A synthetic connection test
An illustrative template. Replace the details with material you are allowed to share.
Test project: Paper Kite Test detail: The fictional prototype uses a green square on its cover. In the second connected assistant, ask: Consult Dray for the Paper Kite test detail. State what you found and its source. If you cannot retrieve it, report that rather than guessing.
USING THIS WITH DRAY
Keep the context. Check the source.
Dray separates web-history capture from stored-context access. CLI agents need their own MCP setup; connecting one does not capture its local files or transcripts.
Dray is in beta. Available sources and connections depend on your setup. Read the data notes before granting access.
Sources & next reads
Product details are grounded in the guides below. Example projects and prompts are illustrative, not customer results.