Memcp

🚀 MemCP - Memory Context Protocol in AI Agent Memory Protocol

MemCP is a modular and extensible AI agent memory MCP server designed for IDEs or LLM client applications that support MCP. It extends the mcp-server example in Zep AI's Graphiti, enabling AI agents to build, maintain, and query a time-series knowledge graph that evolves automatically with new information.

Currently, MemCP is designed as a plug-and-play memory server suitable for any IDE or LLM client application that supports local SSE or standard input/output (stdio) transmission.

⚠️ Important Note

It has been primarily tested with the Cursor IDE and may behave differently compared to other MCP-supporting clients.

✨ Features

• Time-series Knowledge Graph: Allows AI agents to automatically build and query a knowledge graph that evolves over time.
• MCP Integration: Compatible with any MCP-supporting client.
• Entity Extraction: Automatically extracts code-specific entities and their relationships. Customization features are coming soon.
• Flexible Configuration: Can be configured via environment variables, TOML files, or command-line parameters.
• Persistent Storage: Uses Neo4j to store knowledge, enabling cross-session persistence.
• Multi-transport Support: Supports both SSE (HTTP) and stdio transmissions.

📚 Documentation

The complete MemCP documentation is located in the docs directory. Please refer to docs/README.md for more information.

📦 Installation

Requirements

• Operating System: Linux, macOS, or Windows (Linux is recommended).
• Programming Language: Python 3.8+
• Dependency Management Tool: pip (recommended)

Installation Steps

1. Clone the repository:

git clone https://github.com/your-repository/memcp.git
cd memcp

2. Install dependencies:

pip install -r requirements.txt

3. Configure environment variables (optional):
   • Set NEO4J_URI to specify the database connection address.
   • Set NEO4J_USER and NEO4J_PASSWORD for authentication.
   • Set OPENAI_API_KEY or ANTHROPIC_API_KEY for entity extraction.

Start the service

python3 memcp_server.py

🚀 Quick Start

Basic Usage

1. Initialize a MemCP instance:

from memcp import MemCP
mem = MemCP()

2. Add a knowledge snippet:

mem.add_fact("User Preference", "Likes red")

3. Query the knowledge graph:

print(mem.get_facts())

Advanced Usage

Entity Types

MemCP defines the following default entity types:

• Preference: User preferences and inclinations.
• Procedure: Steps or operations for performing specific tasks.
• Requirement: Functional requirements that a product or service must meet.

You can enable/disable these entity types using the --graph.use_memcp_entities flag.

Persistent Storage

MemCP uses the Neo4j database for data persistence. By default, data is automatically saved to the local instance. To configure a custom database address, set the environment variable:

export NEO4J_URI=bolt://localhost:7687

🤝 Contribution Guidelines

Contributions are welcome! Please refer to CONTRIBUTING.md for specific guidelines.

📄 License

MemCP is licensed under the MIT License. Please refer to the LICENSE file for details.

📌 Final Note

⚠️ Important Note

MemCP has a high API call frequency because the Graphiti it depends on also does. This means that while it can generate accurate and useful knowledge graphs, it also incurs high call costs. Please control the number of calls when using it to avoid excessive fees.