mcp_mirror_g0t4_mcp_server_memory_file
by MCP-MirrorMemory Management Server for Claude
Overview
The Memory Management Server for Claude is an implementation of the Model Context Protocol (MCP) designed to help Claude (and other MCP clients) manage inter-chat context using a memory text file. Each line in the file represents a memory, and the server provides tools to add, search, delete, and list memories during a chat session.
Tools
The server offers the following tools for memory management:
memory_add(memory: string): Appends a new memory to the file.memory_search(query: string): Returns memories that match the query (substring exact match).memory_delete(query: string): Deletes memories that match the query (substring exact match).memory_list(): Returns all memories stored in the file.
For example:
- Mentioning a name could add "talking to Wes".
- Mentioning a daughter's age could add "Wes's daughter is 8".
- Correcting a name could involve searching for the old name, deleting it, and adding the new name.
Design
The server uses a simple memory text file for several reasons:
- Inspired by ChatGPT's memory system, which presents memories as a text file.
- Unstructured text simplifies tooling and parameters, reducing complexity.
- Testing with mcp-server-commands showed success when Claude had access to reminders.
Cueing Mechanism
- Training: Models are trained to know when to store memories, similar to tool use.
- Prompt: System prompts likely include reminders to trigger memory storage.
- Tool Alone: Without additional cues, Claude may not store memories, indicating the need for prompt-based reminders.
Usage
The server is particularly useful for maintaining context across chat sessions. When a new chat begins, Claude can automatically retrieve recent memories or request specific ones. These memories can then influence responses, tool usage, and other interactions.
Example Workflow
- Add Memories: Use
memory_addto store relevant information during a chat. - Search Memories: Use
memory_searchto retrieve specific memories based on a query. - Delete Memories: Use
memory_deleteto remove outdated or incorrect information. - List Memories: Use
memory_listto view all stored memories.
TODOs/Ideas
Potential future enhancements include:
- Recency Factor: Rearrange memories based on recency to prioritize recent information.
- Fade Out Old Memories: Gradually deprecate older memories to keep the file relevant.
About
This project is a mirror of g0t4/mcp-server-memory-file.
Resources
Stats
- Stars: 1
- Watchers: 0
- Forks: 0
Languages
- TypeScript: 53.7%
- JavaScript: 46.3%
Invite Gmail Contacts
