Most "connect your agent" setup flows stop too early. A green dot tells you a configuration exists. It does not prove that one agent can retrieve what another agent learned.
Here is the two-minute test I use after connecting any shared-memory service.
- In your first coding agent, store one harmless note:
Store a memory titled "setup check" with the content "shared memory is wired up."
Open a fresh session or a different connected agent.
Ask it:
Recall the memory titled "setup check."
There are only two useful outcomes.
- It returns the note. Good: the connection, account, project scope, and recall path are all behaving together.
- It cannot find the note. Do not keep coding and hope. Check that both clients signed in to the same account and project, then restart the client that was connected before the configuration changed.
This test sounds almost too small, but it catches the failures that matter: a local-only configuration, an expired sign-in, a client connected to the wrong project, or a setup that did not survive restart.
It also gives a team a shared definition of "working." Working is not a dashboard badge. Working means useful context makes it across a real boundary: agent to agent, or session to session.
Once the check passes, replace the test note with five things you are tired of re-explaining: a deployment quirk, a flaky test, an architecture decision, a setup trap, and the convention that keeps coming up in review.
That is enough to make the next session noticeably less blank.
I built StremAI around this exact round-trip. It connects through MCP and lets connected agents store and recall project memory. You can try a recall in the StremAI playground before committing to a workflow.
Top comments (0)