Connect AI agents
Generic MCP adapter
Use the project-neutral MCP integration pattern.
On this page
Use this adapter when an agent has Zharwing Memory MCP tools.
The daily-memory surface is complete and intentionally limited to eleven tools. Administrative daemon methods are not missing MCP capabilities; they belong to the UI and CLI administration interfaces.
Startup#
- Call
memory.get_startup_stateonce for the work round with the current working directory or project id. - Use its compact active/latest/recent summaries to carry forward unfinished tasks, current next steps, current blockers, and recent touched files.
- Do not call
memory.get_latest_sessionwhen startup already supplies enough context. - Call
memory.start_sessionfor a new daily/work-round session by default. - Call
memory.searchwith the task, feature, error, or file names involved. - Search before requesting detail. Call
memory.get_session_detailonly for explicitly needed body or paginated checkpoint history. - Preview or persist a context bundle only when compact state and targeted search are insufficient.
During Work#
- Use
memory.save_checkpointafter meaningful progress. - Supplied
nextStepsandblockersreplace current state; empty arrays clear it; omitted fields preserve it. - Treat sessions, file paths, and routine metadata in the selected project as normal AI-visible memory.
- Use the UI or CLI for project administration, imports, graph settings, backups, Trash, or durable document editing.
Closeout#
Call memory.close_session with:
- concrete summary
- next steps
- blockers, if any
- touched files, if known
The MCP surface is intentionally limited to the daily coding-memory loop.