@agentmemory/mcp

What it does

agentmemory provides persistent memory for AI coding agents through an MCP server interface. Agents can store, query, and retrieve notes about projects, decisions, and learned patterns. The system uses hybrid search with confidence scoring and knowledge graphs to surface relevant memories when agents ask questions. The server acts as an MCP shim, re-exposing the agentmemory library to work with Claude Code, Copilot CLI, Cursor, Gemini CLI, and other MCP-compatible agents.

Who it's for

Developers using AI coding agents (Claude Code, Copilot CLI, Cursor, Gemini CLI) who need their agent to retain project-specific context across sessions. Useful for teams maintaining complex codebases or running multiple agents that require consistent understanding of architecture and conventions.

Common use cases

• Store architectural decisions and project conventions so agents apply them consistently across sessions.
• Avoid repeating the same context and constraints to your agent—let it learn once and retain knowledge.
• Build a searchable knowledge graph of codebase patterns, conventions, and learned workarounds.
• Sync memory across multiple agents working on the same project for consistent understanding.
• Query past decisions to answer "what did we learn about this" questions without re-explaining.

Setup pitfalls

• Four secrets detected in the package—review credentials handling and configuration before deployment.
• Requires filesystem write access to persist memory—ensure proper permissions and sandbox the MCP server appropriately.
• Windows support limited to WSL2; native Windows setup requires manual steps and lacks full agent integration.
• The server makes network calls; verify proxy settings and connectivity if behind a firewall.