memcp

Overview

MemCP is a sophisticated MCP server that gives Claude Code a persistent external memory system. It solves the context loss problem that occurs after /compact commands and when sessions hit context window limits. The server implements 24 MCP tools organized across memory management, context handling, search, graph navigation, and lifecycle management categories. It uses SQLite for storage with a MAGMA 4-graph system that connects insights via semantic, temporal, causal, and entity edges. The architecture follows a 3-layer delegation pattern with comprehensive test coverage and benchmarks showing significant token efficiency advantages over native context-only operation.

Try asking AI

After installing, here are 5 things you can ask your AI assistant:

• memcp_ping

  Health check — returns server status and memory statistics

  Try:Check memory server statusTry:Are you working?
• memcp_remember

  Save an insight to persistent memory

  Try:Remember that we decided to use React for this projectTry:Store this finding about database optimization
• memcp_recall

  Retrieve insights from memory with various filters

  Try:Find insights about authenticationTry:What decisions did we make last week?
• memcp_forget

  Remove an insight from memory by ID

  Try:Forget insight ID 123Try:Delete that old decision about microservices
• memcp_load_context

  Store content as a named context variable on disk

  Try:Save this file as project_requirementsTry:Load the API documentation as api_docs
• memcp_inspect_context

  Inspect a stored context's metadata and preview

  Try:Show me what's in project_requirementsTry:Check the size of api_docs
• memcp_get_context

  Read a stored context's content or specific line range

  Try:Read project_requirementsTry:Get lines 10-20 from api_docs
• memcp_chunk_context

  Split a stored context into navigable numbered chunks

  Try:Chunk the requirements document by paragraphsTry:Split api_docs by headings
• memcp_peek_chunk

  Read a specific chunk from a chunked context

  Try:Show me chunk 3 of project_requirementsTry:What's in chunk 5 of the documentation?
• memcp_search

  Search across memory insights and context chunks

  Try:Search for JWT authenticationTry:Find all mentions of payment processing
• memcp_related

  Find connected knowledge via graph edges from an insight

  Try:What's related to insight 456?Try:Find things connected to the database decision
• memcp_projects

  List all projects with insight, context, and session counts

  Try:Show all projectsTry:What projects have we worked on?

Comparable tools

Installation

pip install memcp

bash scripts/install.sh

{
  "mcpServers": {
    "memcp": {
      "command": "python",
      "args": ["-m", "memcp.server"],
      "env": {}
    }
  }
}

pip install memcp[ner]  # For spaCy NER
pip install memcp[vector]  # For semantic search with embeddings

FAQ

How does MemCP prevent context loss after /compact?

MemCP implements auto-save hooks that force Claude to persist insights before compact, storing them in SQLite for retrieval across sessions.

What file storage does MemCP use?

MemCP uses SQLite for the knowledge graph (graph.db) and filesystem for contexts and chunks (~/.memcp/).

Compare memcp with