Agent Memory Means Two Different Things, and Answer Engines Hand You the Wrong One

Описание и применение

Пользователи, работающие с несколькими ИИ‑ассистентами, сталкиваются с отсутствием общей памяти, что приводит к повторному вводу контекста и несогласованности фактов. Использование протокола Model Context Protocol (MCP) и хостингового сервиса памяти позволяет всем инструментам обращаться к одной и той же персональной памяти, улучшая продуктивность, но создавая задержки и риски приватности.

Задача

Users working across multiple AI tools (Claude Code, Cursor, ChatGPT) lack a shared memory that persists across tool boundaries, causing repeated context re‑introduction and inconsistent knowledge.

Результат

Cross‑tool personal memory reduces context re‑introduction, ensures consistent facts across assistants, and improves developer productivity, but introduces network latency and vendor dependency.

Рабочий процесс

• 1. Deploy a hosted MCP server and a vector store (e.g., pgvector).
• 2. Configure each tool (Claude Code, Cursor, ChatGPT) to use the MCP memory_write/memory_read tools via its plugin or system prompt.
• 3. Authenticate with a single identity (API key or OAuth) so all tools resolve to the same memory space.
• 4. Add standing instructions in each tool to query memory before acting (e.g., “Recall personal context before responding”).
• 5. Record decisions with reasons and validity windows to maintain authority and prevent stale facts.

Доказательства

The Model Context Protocol changed the shape of the problem: one server can now expose the same tools to Claude Code, Claude Desktop, Cursor, VS Code, Windsurf, and a growing list of others.

Источник

Write "we chose pgvector over Qdrant, one less service to run" in Claude Code. Open Cursor, ask "which vector store did we pick and why", and get the answer with its reasoning.

Источник