memory-mcp

cstax-elire/memory-mcp

3.1

If you are the rightful owner of memory-mcp and would like to certify it and/or have it hosted online, please leave a comment on the right or send an email to dayong@mcphub.com.

The Memory MCP (Postgres) is a standalone stdio JSON-RPC server designed to interface with a Postgres database, providing tools for data retrieval and write-back operations.

Tools
4
Resources
0
Prompts
0

Memory MCP (Postgres)

Overview

  • Standalone stdio JSON-RPC MCP server that connects to your Postgres “brain”.
  • Exposes tools for retrieval (context_pack, search, conversation) and write‑back (nuggets, facts, summaries, entities, relationships, tasks/goals).
  • Designed to be global: place this folder at ~/Documents/Projects/memory-mcp and run the command from anywhere.

Install

  • Copy this folder to: ~/Documents/Projects/memory-mcp
  • Install deps:
    • cd ~/Documents/Projects/memory-mcp
    • pip install -r requirements.txt
  • Configure env:
    • Option A: Project-local → copy .env.example to .env and edit DB_* and (optional) OPENAI_API_KEY
    • Option B: Global → create ~/.memory-mcp.env with DB_* and OPENAI_API_KEY

Run (global)

  • From any directory:
    • ~/Documents/Projects/memory-mcp/bin/memory-mcp

Claude Code Integration

  • Set the memory tool command to: ~/Documents/Projects/memory-mcp/bin/memory-mcp
  • Health checks:
    • echo '{"id":"1","method":"initialize"}' | ~/Documents/Projects/memory-mcp/bin/memory-mcp
    • echo '{"id":"2","method":"tools/list"}' | ~/Documents/Projects/memory-mcp/bin/memory-mcp

Core Tools

  • context_pack(query[, final_n, tags, since, until, entity_name]) → fused high‑signal context with citations
  • search(query, kind=messages|summaries|nuggets|facts[, tags, since, until, top_k]) → targeted lookups
  • conversation(item_id[, limit, offset] | [center, window]) → paginated/windowed transcript
  • stats() → basic counts

Write‑Back Tools

  • insert_nugget, insert_fact, upsert_summary, classify, relate_conversations, evolve_concept
  • upsert_entity, link_entity, create_task, update_task, tasks_for, goals
  • morning_brief([since_date, horizon_days, limit, tags, entity_name]) → new nuggets/facts + tasks due/blocked
  • outline([query, since_date, until_date, tags, entity_name, limit]) → date‑ordered outline with counts + preview

Docs

  • See docs/MCP_INTEGRATION.md for API usage and examples
  • See docs/MEMORY_AUTOPILOT.md to guide Claude when to retrieve and how to write back

Notes

  • OPENAI_API_KEY is optional; without it, full‑text search still works. With it, vector ranking improves.
  • The server is stateless; Claude spawns it on demand. No daemon required.