Use learning and memory commands
Use these commands to connect protected workspace services and maintain stored learning. They appear only when the connected server supports them, so an embedded session and a remote server can expose different commands.
Workspace service enrollment
When the server provides protected workspace services, run /mcp-refresh to
start or recheck the connection. This broker path keeps the existing destructive
reconnection disclosure and browser consent flow. /tools-connect is a
deprecated broker-only alias. Run /tools-cancel to cancel a pending connection.
You can continue editing the prompt while browser consent is pending. If the
server requires the connection before accepting a prompt, mecatui sends the
retained prompt after the connection succeeds.
On a direct MCP server, /mcp-refresh performs the direct refresh without a
browser consent flow. mecatui enables exactly one path from server
capabilities. It refuses the command when both modes, neither mode, or the
matching client collaborator are unavailable.
Completed-session learning
| Command | Available values | Effect |
|---|---|---|
/learning | Off, Review, Auto | Sets how the server handles learning from completed sessions. |
/learning-sensitivity | Conservative, Balanced, Eager | Sets the evidence threshold for learning proposals. |
Both commands change pending settings and require a server restart. They do not change a separately configured consolidation schedule.
In embedded mode, these settings live with the embedded server. In connect
mode, ask the remote server operator to change the settings and restart the
server. For configuration details, see
Memory and knowledge.
Review and maintain memory
/dreamproposes maintenance for the selected project-memory or user-model store. Generating a plan sends selected memory values to the configured model and uses tokens. Review the plan before applying it. Confirmation applies the whole plan; dismissal changes nothing. A server restart discards pending plans./reflectionsopens staged learning proposals. Review their evidence and approve or reject each proposal when the server allows it./reflectsubmits the completed session for immediate reflection, even when automatic learning is off. On completion, it reports a proposal count, an abstention because the evidence or bounds were insufficient, or a failure. An abstention does not stage a proposal.
If a command is absent or unavailable, the connected server does not support it under its current configuration and policy. For learning, consolidation, and authorization behavior, see Learning and Dreaming and memory consolidation. Builders configuring stores, retention, or learning should use the memory guide.