A Mem Mcp
Give coding agents persistent, self-updating memory across sessions without rebuilding context from scratch each time.
Overview
a-mem-mcp is a MCP server for the Build phase that gives AI coding agents a self-evolving, ChromaDB-backed memory layer with pluggable LLM backends.
What is this MCP server?
- Self-evolving memory store backed by ChromaDB with configurable local path (CHROMA_DB_PATH)
- Embedding pipeline via sentence-transformers (default all-MiniLM-L6-v2)
- LLM backends: openai, ollama, sglang, or openrouter with LLM_MODEL selection
- stdio PyPI package identifier a-mem at version 0.2.1
- OpenAI API key required when LLM_BACKEND=openai for memory consolidation
- Server version 0.2.1 on PyPI identifier a-mem
- 4 LLM_BACKEND options: openai, ollama, sglang, openrouter
- Default embedding model all-MiniLM-L6-v2
Community signal: 34 GitHub stars.
What problem does it solve?
Agents forget what you already decided, so you re-explain architecture, naming, and constraints on every new chat.
Who is it for?
Indie builders running long-horizon agent workflows who want local vector storage and a choice of OpenAI or self-hosted LLM backends.
Skip if: Teams that only need a single-session transcript or already centralize memory in a hosted product with its own MCP.
What do I get? / Deliverables
After you register the stdio server and set API keys, the agent can store and retrieve evolving memories so multi-session work stays aligned.
- Registered stdio MCP server in your agent config
- Persistent ChromaDB-backed memory the agent can read and update
- Configurable embedding and LLM backend for memory operations
Recommended MCP Servers
Journey fit
Agent memory sits in the build phase because it is infrastructure you wire into Claude Code, Cursor, or Codex while assembling your agent stack. agent-tooling is the canonical shelf for MCP servers that extend what agents can remember and recall during implementation work.
How it compares
MCP memory integration, not a one-off brainstorming or planning skill.
Common Questions / FAQ
Who is a-mem-mcp for?
Solo and small-team builders using Claude Code, Cursor, or Codex who need persistent, updating agent memory across coding sessions.
When should I use a-mem-mcp?
Use it while building your agent stack when repeated context loss is slowing implementation and you are ready to configure ChromaDB plus an LLM backend.
How do I add a-mem-mcp to my agent?
Install the PyPI package a-mem, set transport stdio in your MCP config, and provide OPENAI_API_KEY (if using OpenAI), LLM_BACKEND, LLM_MODEL, EMBEDDING_MODEL, and CHROMA_DB_PATH.