agent-memory-mcp

Agent Memory MCP - Persistent Memory System for AI Agents

Overview of Features

Agent Memory MCP is a hybrid memory system designed for AI agents, providing persistent, searchable knowledge management capabilities. It supports storage and retrieval of long-term knowledge such as architectural patterns and technical decisions.

Use Cases

AI Agent Development

When your agent needs to remember project architecture, technology choices, or development decisions across sessions, Agent Memory MCP provides dedicated memory storage and retrieval capabilities, preventing the agent from repeatedly forgetting critical context.

Team Knowledge Retention

Record architecture decision records (ADR), design pattern selections, code conventions, and other content. A searchable memory store allows team members and agents to quickly access historical decision context.

Knowledge Base and Agent Integration

Sync project documentation and technical specifications with the agent workspace. The agent can quickly retrieve relevant knowledge via memory_search, improving the accuracy and consistency of responses.

Core Features

Searchable Memory Store (memory_search)

Supports searching by query keywords, type (e.g., pattern, decision), or tags, helping agents quickly locate historical knowledge. For example, search all "authentication patterns" or architecture decisions with a specific tag.

Knowledge Write and Classification (memory_write)

When recording new knowledge or decisions, you can specify a key, type, and tags for easier subsequent categorization. Suitable for storing architectural decision records, design pattern descriptions, technical selection rationales, and other structured knowledge.

Standalone Visual Dashboard

Built-in web dashboard (http://localhost:3333) visually displays memory usage statistics and content distribution, helping developers understand the accumulation of the knowledge base and the agent's "learning" progress.

Frequently Asked Questions

What is Agent Memory MCP?

Agent Memory MCP is a memory server for agents based on the MCP protocol, providing persistent knowledge management capabilities for AI agents. Through tools like memory_search, memory_write, and memory_read, it enables agents to remember and retrieve architectural decisions, design patterns, technical conventions, and other long-term knowledge across sessions.

How do I install and configure agent-memory-mcp?

Installation requires Node.js v18 or higher. First, clone the repository into the .agent/skills/agent-memory directory, run npm install and npm run compile to build the project. Then start the MCP server and connect it to your project workspace with npm run start-server <project_id> <workspace_path>.

Where is the memory data stored?

Memory data is stored at the specified location in the target workspace and is accessed via the MCP server for read/write operations. Data is saved in a structured form and supports three retrieval methods: exact queries by key, full-text search, and filtering by type or tags. You can use the memory_stats tool to view memory store usage statistics.