What is Memory Server

Memory Server is a semantic memory storage and retrieval system designed to work with the Model Context Protocol (MCP). It utilizes Quarkus for performance, Kotlin for type safety, and Qdrant for efficient vector-based semantic search to provide high-speed memory management.

Use cases

Memory Server can be utilized in various AI applications where semantic understanding of information is necessary. It is particularly useful for agents that require memory capabilities for tasks such as question answering, recommendations, and contextual assistance, facilitating effective information retrieval based on meaning rather than keywords.

How to use

To begin using Memory Server, you can run it locally with Docker Compose. For memory operations, the command-line interface (CLI) can be employed to store (‘remember’) and fetch (‘recall’) memories based on semantic content. Integration with AI agents like Goose is achieved by using the memory-proxy extension to facilitate communication via MCP calls.

Key features

Key features include semantic vector memory for nuanced recall, tag-based filtering for organizing memories, compatibility with MCP for integrating with AI agents, and providing both a CLI for human operators and an API for agent interaction, supporting flexible and scalable memory operations.

Where to use

Memory Server is suitable for deployments in AI-driven environments that demand advanced semantic memory capabilities. It can be applied in chatbots, virtual assistants, customer support systems, and any application involving natural language processing where context awareness and memory are essential for enhanced user interactions.

Memory Server

Qdrant × Quarkus MCP Memory Server
High-performance semantic memory orchestration for agents and daemons.

Overview

Memory Server provides a powerful semantic memory storage and retrieval system tailored specifically for use with the Model Context Protocol (MCP). It leverages the speed of native Quarkus applications, Kotlin for type-safety, and Qdrant for scalable vector-based semantic search.

Features

• Semantic Vector Memory: Store and recall memories based on semantic similarity rather than exact matches.
• Tag-based Filtering: Easily organize and filter memories using flexible tagging.
• Dual CLI and Agent Interface: Designed for both human operators (memory-cli) and AI agents (memory-proxy).
• MCP Compatible: Uses standard MCP calls for seamless integration with AI agents like Goose.

Quickstart

Local Development

To run Memory Server locally using Docker Compose:

docker-compose up

CLI Usage

To use the command-line client for memory operations:

./memory-cli remember --content "This is a test memory" --tags "qa:true,env:dev"
./memory-cli recall --content "test memory"

Agent (Goose) Integration

Load the memory-server extension via Goose CLI with STDIO proxy:

/extension memory-proxy

Example MCP calls via Goose:

{
  "request": {
    "memory": {
      "content": "Semantic search test"
    },
    "page": 1,
    "pageSize": 10
  }
}

Architecture

memory-cli ↔ memory-proxy ↔ REST (memory-server) ↔ gRPC ↔ Qdrant

• memory-server: REST endpoint, semantic embedding generation, gRPC client to Qdrant.
• memory-cli: User/operator CLI.
• memory-proxy: STDIO proxy enabling MCP tool calls from AI agents like Goose.

Recipes & Usage

• QA Recipes available in recipes/.
• Example configuration and recipes provided for immediate integration with Goose CLI.

Contributing

Contributions are welcome! Please open an issue or pull request on GitHub.

License

This project is licensed under the APACHE 2.0 License – see the LICENSE file for details.