cdeust/Cortex

Overview

Cortex provides persistent memory for Claude Code using PostgreSQL and pgvector. It stores and retrieves information across sessions based on 26 biological mechanisms with ablation evidence from 41 neuroscience papers, achieving 98.4% R@10 on LongMemEval and 94.2% on LoCoMo.

Best for

Best for
Developers who need reliable long-term memory for Claude Code in complex, multi-session projects.

Use cases

• Maintaining conversation context across long Claude Code sessions
• Retrieving relevant prior knowledge during agentic coding tasks
• Building memory-augmented coding assistants with verifiable retrieval

How to use

Install

uvx hypermnesia-mcp          # run the MCP server directly

Tools exposed

• Open-domain
• Multi-hop
• Single-hop

Tested with

Claude Desktop, Claude Code

Notes

Cortex provides persistent memory for Claude Code using PostgreSQL and pgvector. It stores and retrieves information across sessions based on 26 biological mechanisms with ablation evidence from 41 neuroscience papers, achieving 98.4% R@10 on LongMemEval and 94.2% on LoCoMo.

41 stars on GitHub. Last updated 2026-06-01.

Use cases

• Maintaining conversation context across long Claude Code sessions
• Retrieving relevant prior knowledge during agentic coding tasks
• Building memory-augmented coding assistants with verifiable retrieval

Pros

• High retrieval accuracy validated on multiple benchmarks (LongMemEval, LoCoMo, BEAM-10M)
• Neuroscience-inspired design with per-mechanism ablation evidence for transparency
• Uses widely compatible PostgreSQL with pgvector extension

Cons

• Requires managing a PostgreSQL instance with pgvector
• Tied specifically to Claude Code, limiting reuse with other LLMs
• Setup and tuning may be non-trivial for casual users

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