MemCP - Memory Context Protocol for AI Agents

Name: MemCP - Memory Context Protocol for AI Agents
Author: evanmschultz

@evanmschultz

Repository

About MemCP - Memory Context Protocol for AI Agents

A memory MCP for IDE AI agent context support based on the mcp_server example by Zep AI

Basic information

Config

Add this server to your MCP-compatible client using the configuration below.

{
  "mcpServers": {
    "memcp": {
      "command": "uv",
      "args": [
        "sync"
      ]
    }
  }
}

Tools

Add an episode to the knowledge graph (text, JSON, messages)

Search for entity nodes in the graph

Search for relationships between entities

Delete a relationship between entities

Delete an episode from the knowledge graph

Get details about a specific relationship

Retrieve recent episodes

Reset the knowledge graph (use with caution)

Overview

What is MemCP?

MemCP is an extensible memory MCP server for AI coding agents that builds, maintains, and queries a temporal knowledge graph of facts, entities, and relationships. It modularizes and extends the Zep AI Graphiti MCP server and works with any MCP-compatible IDE or LLM client app through local SSE or stdio connections.

How to use MemCP?

Install prerequisites (Python 3.10+, Neo4j 5.26+, OpenAI API key), clone the repository, install using uv sync or pip install memcp, copy and edit the .env file, then run memcp to start the server. For Cursor IDE, add a configuration block with "transport": "sse" and "url": "http://localhost:8000/sse" to the MCP servers settings.

Key features of MemCP

Temporal knowledge graph that evolves with new information
Automatic entity and relationship extraction
Works with any MCP-compatible client
Flexible configuration via env vars, TOML, or CLI
Persists knowledge in Neo4j across sessions
Supports both SSE and stdio transports

Use cases of MemCP

AI coding agents that recall user preferences across sessions
Maintaining a running context of project requirements and procedures
Building a searchable memory of past episodes for agent decision-making
Enabling multi-session continuity for LLM-based development tools

FAQ from MemCP

What runtime dependencies does MemCP require?

MemCP requires Python 3.10+, a Neo4j database (version 5.26 or later), and an OpenAI API key. An Anthropic API key is optional for using Claude models.

How does MemCP store data?

MemCP stores knowledge in a Neo4j database for persistence across sessions. You can run Neo4j with Docker if you prefer not to install it locally.

What transports does MemCP support?

MemCP supports both SSE (HTTP) and stdio transports for integration with different MCP-compatible clients.

Are there any known limitations?

MemCP makes a large number of API calls, which incur significant costs. It has been tested primarily with the Cursor IDE, and use with other clients may require additional configuration.

How is MemCP configured?

MemCP can be configured via command-line arguments (highest priority), a config.toml file, or default values (lowest priority). Environment variables from a .env file are automatically loaded on startup.