Giving every AI tool the same memory (Claude, ChatGPT, Cursor, Codex, Gemini)
- 发布
- 09/05 00:51
- 收录
- 09/05 01:00
- 来源类型
- 开发者社区
- 档位
- 社区
- 信源状态
- 同步延迟
讨论趋势
百分比基于采集到的讨论信号，不代表新增评论数或独立参与人数。曲线仅用于同一话题在不同时段的比较。
开发者在使用Claude、ChatGPT、Cursor、Codex和Gemini等AI编码工具时，常因工具间缺乏共享记忆而感到沮丧。为解决此问题，有人提出将记忆层作为MCP服务器运行。这样，每当工具处理请求时，都可以查询MCP以获取相关上下文并将其包含在提示中，并在处理后将任何新的相关上下文提交给MCP。由于它以MCP形式实现，任何能够读写MCP的工具都能理解工作的完整上下文，从而省去了反复解释的麻烦，但需要注意的是，这会增加每次请求的延迟。
If you're jumping between different coding tools during the day, you'll know how frustrating it can be to solve something in Claude, then hop to ChatGPT and have it not understand a thing.
The solution is to run a memory layer as an MCP server that all your tools can tap into - every time they process a request, they query the MCP for relevant context to include in the prompt, and submit any new relevant context to the MCP after the fact. Since it's implemented as an MCP, any tool that can read/write to an MCP (Claude, ChatGPT, Cursor, Codex, Gemini, etc) can understand the full context of your work without you having to painstakingly explain things.
A few things that are important to consider if you're looking to implement or evaluate one:
- Semantic search is important, since keyword matching will miss paraphrased context
- Tools should be able to define what information to share with whom
- Latency is a consideration, as this will add overhead to every request