AGENTS.md

Guidance for coding agents working on LMSSEP Memory MCP Server. This server provides long-term conversation memory backed by ChromaDB.

Project overview

The Memory server stores conversations with embeddings, supports semantic query, updates existing entries, deletes entries, retrieves individual conversations, lists stored conversations, searches by time, and can clear stored memory. Its persistent database is generated runtime data, not source.

Setup commands

bashcd <project-root>/genesis/LMSSEP/memory
pip install -r requirements.txt
python3 main.py

Configuration lives in configs/memory.py, including enablement, ChromaDB parameters, and embedding settings.

Code and data map

• main.py starts the MCP server.
• requirements.txt includes FastMCP and ChromaDB dependencies.
• configs/memory.py controls memory behavior.
• docs/memory-tools.md documents tool contracts.
• output/chroma_db/ stores generated vector database files.

Tool usage rules

Use conversation_memory_store only for conversations that should be remembered. Use concise summaries and metadata that avoid secrets. Use conversation_memory_query for retrieval, conversation_memory_update for corrections, and delete or clear tools only with explicit user intent.

Testing and verification

Create a small synthetic memory record, query it by related wording, fetch it by id, update its summary, then delete it. Use placeholder conversation ids in public notes and remove the synthetic record when done.

Security and privacy

Conversation memory can retain sensitive user data. Never store credentials, tokens, private keys, personal identifiers, or raw incident transcripts unless the user explicitly asked for persistent storage and policy allows it. Keep output/chroma_db/ out of commits and public artifacts.

Commit and PR rules

Changes that affect embedding, retention, deletion, or clear behavior need focused tests and an explicit privacy note in the PR. Do not commit generated ChromaDB files.