Pre-submission checklist | 提交前检查
Problem Statement | 问题陈述
Hi MemOS team,
I’d like to propose an official Claude Code integration for MemOS.
Background
Recently, tools like claude-mem have become popular, which shows that the demand for long-term memory in Claude Code is very real. Developers want Claude Code to remember project context, user preferences, historical conclusions, and reusable workflows across sessions.
However, the current third-party solutions usually have several limitations:
① They tend to be local-only and isolated
② They may consume extra memory or require always-on background processes
③ They are hard to govern as history grows
④ They do not naturally share memory across clients, devices, or other agents
Why MemOS is a better fit
I think MemOS is a much better foundation for this than a standalone memory patch, because MemOS can provide:
① Structured memory storage instead of simple local caching
② Semantic retrieval, deduplication, reranking, and lifecycle management
③ Better handling of conflicting old/new information
④ Optional local-first or cloud-backed deployment
⑤ Reusable experience crystallization into Skills
⑥ Cross-client reuse across Claude Code, Cursor, Hermes, OpenClaw, and other agent environments
Suggested direction
Instead of building a heavy native integration from day one, I suggest starting with an official Claude Code MCP integration or lightweight plugin layer.
A good minimum viable version could include:
① search_memory before answering
② add_message after answering
③ Project-level isolation
④ User-level long-term continuity across sessions
⑤ Optional auto-injection switch
⑥ Support for both local and cloud modes
Expected value
This would make MemOS not just “another memory helper”, but a reusable memory layer for coding agents.
Compared with standalone tools, MemOS could offer:
① More controllable memory behavior
② Better long-term maintainability
③ Lower risk of memory pollution
④ Shared memory infrastructure across multiple agent products
⑤ A path from memory to reusable skill
Design principles
My suggestion is:
① Keep it lightweight
② Do not deeply modify Claude Code core behavior
③ Make auto-injection optional
④ Prioritize controllability over excessive automation
⑤ Clearly separate rules, memory, and knowledge base usage
Potential future enhancements
After the basic integration works well, future versions could add:
① One-click installer
② Project binding
③ Viewer quick entry
④ Debugging/logging panel
⑤ Knowledge base association
⑥ Skill evolution based on repeated successful coding workflows
Why this matters
Claude Code is becoming a very important agent entry point for developers. If MemOS provides a clean and reliable integration here, it could become one of the strongest developer-facing entry points in the MemOS ecosystem.
Thanks for considering this feature request.
Willingness to Implement | 实现意愿
Pre-submission checklist | 提交前检查
Problem Statement | 问题陈述
Hi MemOS team,
I’d like to propose an official Claude Code integration for MemOS.
Background
Recently, tools like claude-mem have become popular, which shows that the demand for long-term memory in Claude Code is very real. Developers want Claude Code to remember project context, user preferences, historical conclusions, and reusable workflows across sessions.
However, the current third-party solutions usually have several limitations:
① They tend to be local-only and isolated
② They may consume extra memory or require always-on background processes
③ They are hard to govern as history grows
④ They do not naturally share memory across clients, devices, or other agents
Why MemOS is a better fit
I think MemOS is a much better foundation for this than a standalone memory patch, because MemOS can provide:
① Structured memory storage instead of simple local caching
② Semantic retrieval, deduplication, reranking, and lifecycle management
③ Better handling of conflicting old/new information
④ Optional local-first or cloud-backed deployment
⑤ Reusable experience crystallization into Skills
⑥ Cross-client reuse across Claude Code, Cursor, Hermes, OpenClaw, and other agent environments
Suggested direction
Instead of building a heavy native integration from day one, I suggest starting with an official Claude Code MCP integration or lightweight plugin layer.
A good minimum viable version could include:
① search_memory before answering
② add_message after answering
③ Project-level isolation
④ User-level long-term continuity across sessions
⑤ Optional auto-injection switch
⑥ Support for both local and cloud modes
Expected value
This would make MemOS not just “another memory helper”, but a reusable memory layer for coding agents.
Compared with standalone tools, MemOS could offer:
① More controllable memory behavior
② Better long-term maintainability
③ Lower risk of memory pollution
④ Shared memory infrastructure across multiple agent products
⑤ A path from memory to reusable skill
Design principles
My suggestion is:
① Keep it lightweight
② Do not deeply modify Claude Code core behavior
③ Make auto-injection optional
④ Prioritize controllability over excessive automation
⑤ Clearly separate rules, memory, and knowledge base usage
Potential future enhancements
After the basic integration works well, future versions could add:
① One-click installer
② Project binding
③ Viewer quick entry
④ Debugging/logging panel
⑤ Knowledge base association
⑥ Skill evolution based on repeated successful coding workflows
Why this matters
Claude Code is becoming a very important agent entry point for developers. If MemOS provides a clean and reliable integration here, it could become one of the strongest developer-facing entry points in the MemOS ecosystem.
Thanks for considering this feature request.
Willingness to Implement | 实现意愿