STiFLeR7/memex

Overview

STiFLeR7/memex provides persistent memory for AI coding agents through the Model Context Protocol (MCP). It builds a bitemporal knowledge graph of a codebase using Tree-sitter for parsing and Gemini Flash for embeddings, stored in Neo4j via Graphiti. The tool offers 12 MCP tools, hierarchical clustering, and a two-regime confidence decay system for memory management.

Best for

Best for
Developers using MCP-compatible coding agents who need persistent, temporally aware codebase memory

Use cases

• Give Claude Code or Cursor long-term recall of code structure and changes across sessions
• Query historical codebase context with temporal awareness for debugging or refactoring
• Integrate persistent memory into any MCP-compatible agent workflow

How to use

Install

npx stifler-memex-mcp init --repo .

Tools exposed

• get_project_context
• get_symbol_context
• get_recent_decisions
• get_open_problems
• search_context
• get_stale_context
• explain_change
• predict_impact
• record_decision
• record_problem
• resolve_problem
• invalidate_edge

Tested with

Claude Code, Cursor

Notes

STiFLeR7/memex provides persistent memory for AI coding agents through the Model Context Protocol (MCP). It builds a bitemporal knowledge graph of a codebase using Tree-sitter for parsing and Gemini Flash for embeddings, stored in Neo4j via Graphiti. The tool offers 12 MCP tools, hierarchical clustering, and a two-regime confidence decay system for memory management.

11 stars on GitHub. Last updated 2026-06-01. Licensed MIT.

Use cases

• Give Claude Code or Cursor long-term recall of code structure and changes across sessions
• Query historical codebase context with temporal awareness for debugging or refactoring
• Integrate persistent memory into any MCP-compatible agent workflow

Pros

• Bitemporal graph captures both state and history of code for rich context
• Works with multiple popular coding agents via standard MCP interface
• Confidence decay helps manage stale or less relevant information automatically

Cons

• Requires running a Neo4j database, adding infrastructure overhead
• Depends on Gemini Flash for embeddings, limiting offline or alternative model use
• Small community (11 stars) means limited support and documentation

Indexed from awesome-mcp-servers-punkpeye and enriched against its public facts.