awesome-claude-code Add MAMA - Memory-Augmented MCP Assistant

Resource Information

Resource Name: MAMA (Memory-Augmented MCP Assistant)

Resource URL: https://github.com/jungjaehoon-lifegamez/MAMA

Category: Hooks / Tooling

Description: Always-on companion for Claude that remembers your decisions and their evolution. Local-first memory using SQLite + transformers.js embeddings.

Why it's awesome

Tracks decision evolution: Remembers why you chose it, what you tried before, and what didn't work
Local-first architecture: SQLite database + semantic search (transformers.js embeddings)
Cross-platform: Works for both Claude Code (plugin) and Claude Desktop (MCP server)
Auto-context injection: Uses hooks to automatically surface relevant past decisions
No API calls: Fully local embeddings with no network dependencies

Installation

Claude Code:

/plugin marketplace add jungjaehoon-lifegamez/claude-plugins
/plugin install mama

Claude Desktop:

{
  "mcpServers": {
    "mama": {
      "command": "npx",
      "args": ["-y", "@jungjaehoon/mama-server"]
    }
  }
}

Additional Context

MAMA is a complete decision memory system that goes beyond simple note-taking:

Semantic search across decision history
Decision graph tracking (supersedes, refines, contradicts relationships)
Automatic context injection via UserPromptSubmit and PreToolUse hooks
134 passing tests with 100% reliability

Repository includes comprehensive documentation, developer guides, and architecture deep-dive.

Nov 22 '25 04:11 hoonsam

thanks @hoonsam - i started browsing the codebase, it looks pretty interesting - to be honest, seems very well thought out - def a lot to take in - i'm going to be rolling out an idea for how to help "evaluate" complex frameworks like this. so hopefully that will help me get a better understanding for these types of projects than, i don't know, trying it out for a whole week, for example, which is not maintainable for me. maybe you cover this somewhere in the documentation, but do you have any thoughts on how this relates to a more familiar notion of an architectural design record (ADR)? it seems like maybe this is kind of a graph system for ADR's (?) (amongst other things), which would be pretty cool, but I'm just asking about that topic because it would me get a picture. anyway, looks like nice work, thanks for sharing it.

Nov 22 '25 06:11 hesreallyhim

Hi,

Thanks so much for taking the time to look through the codebase - really appreciate the thoughtful feedback!

You're absolutely right about the ADR connection. That's actually a really good way to think about it. The main difference I've noticed in practice is that traditional ADRs are written for humans to read and then explain to the LLM. But with MAMA, the decisions are captured in a way that the LLM itself actually participated in creating, so there's this trust factor that I didn't expect.

What I mean is - before MAMA, I'd point Claude to past decisions and it would still re-investigate everything "just to be sure." Now it actually builds on the previous reasoning, kind of like how you'd trust a colleague's analysis more than just finding a random doc. The superseded/contradicts/refines relationships help track when our thinking evolves, which happens a lot in real projects.

The efficiency gain turned out to be pretty significant too - dropping from around 700 tokens of re-explaining context down to about 50 when you resume a session. And I recently tested it across Claude Code, Codex, Antigravity IDE, and regular Claude Desktop - they all share the same SQLite database which is nice.

Honestly the session continuity feature (checkpoint/resume) kind of emerged as the killer feature from actual usage. It wasn't even the main focus initially.

Re: your evaluation framework idea - that sounds really interesting. The "how do you maintain understanding of complex systems without spending a week on each one" problem is exactly what led me down this path. I'd be happy to help if MAMA could be useful as a test case or something.

Anyway, getting MAMA into awesome-claude-code would be great for helping people discover it, and also for connecting with others working on similar problems.

Some docs that might help:

Developer Playbook has the architecture decisions:

https://github.com/jungjaehoon-lifegamez/MAMA/blob/main/docs/development/developer-playbook.md

Session Continuity section in the README:

https://github.com/jungjaehoon-lifegamez/MAMA#-session-continuity---never-lose-your-context

Let me know if you have any questions!

Best, Jaehoon

2025년 11월 22일 (토) 오후 3:48, Really Him @.***>님이 작성:

hesreallyhim left a comment (hesreallyhim/awesome-claude-code#331) https://github.com/hesreallyhim/awesome-claude-code/issues/331#issuecomment-3565854217

thanks @hoonsam https://github.com/hoonsam - i started browsing the codebase, it looks pretty interesting - to be honest, seems very well thought out - def a lot to take in - i'm going to be rolling out an idea for how to help "evaluate" complex frameworks like this. so hopefully that will help me get a better understanding for these types of projects than, i don't know, trying it out for a whole week, for example, which is not maintainable for me. maybe you cover this somewhere in the documentation, but do you have any thoughts on how this relates to a more familiar notion of an architectural design record (ADR)? it seems like maybe this is kind of a graph system for ADR's (?) (amongst other things), which would be pretty cool, but I'm just asking about that topic because it would me get a picture. anyway, looks like nice work, thanks for sharing it.

— Reply to this email directly, view it on GitHub https://github.com/hesreallyhim/awesome-claude-code/issues/331#issuecomment-3565854217, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFAL6LOSKZGDWDOMNU2VBXD36ABLDAVCNFSM6AAAAACM335EZWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTKNRVHA2TIMRRG4 . You are receiving this because you were mentioned.Message ID: @.***>

Nov 22 '25 07:11 hoonsam