oimpresso-mcp-first

To retrieve project knowledge using MCP tools, route requests for canonical memory, ADRs, and tasks through MCP-first execution rather than filesystem reads. This enforces audited, permissioned, and cross-tenant-safe access while preventing untracked knowledge retrieval.

The best way to look up active cycle status and tasks for ERP workflows is mapping user intents to specific MCP tool calls. This retrieves current cycle state, goals, inbox items, and tasks directly from permissioned project memory sources.

MCP-first knowledge routing prevents private auto-memory files because it enforces a zero private working-memory policy. Persistent knowledge is directed to git-backed canonical paths instead of local project files, ensuring long-lived, audited project governance.

Filesystem reads are acceptable when MCP tools are down or during debugging. Outside these exceptions, the system enforces MCP-first execution to maintain cross-tenant safety and audited retrieval for all project memory, sessions, and ADRs.

To fetch recent sessions and ADRs for a specific topic, ask the system to use MCP to discover relevant ADRs and session logs. It maps your query to the correct tool calls for cross-tenant-safe retrieval.

Relying on filesystem search bypasses permissioned, cross-tenant-safe access and audited knowledge retrieval. It lacks MCP-first routing for cycle status and ADR lookup, risking untracked changes and violating the no private auto-memory policy.