xgmem MCP Memory Server
xgmem is a TypeScript-based Model Context Protocol (MCP) server for enabling project-specific and knowledge graph-based memory for Claude, LLM agents, and other tools. It supports storing, retrieving, and managing entities, relations, and observations per project, with a focus on flexibility and cross-project knowledge sharing.
Features
- Knowledge graph storage for entities, relations, and observations
- CRUD operations via MCP tools
- Persistence to disk (memory.json)
- Docker and TypeScript support
Use Case
xgmem is ideal for:
- Agents and LLMs that need to store and retrieve structured memory (entities, relations, observations) per project.
- Cross-project knowledge sharing and migration.
- Scalable, disk-persistent, and queryable memory for agent ecosystems.
Usage
MCP Config Example
Add to your MCP config (e.g., for windsurf):
"mcpServers": {
"xgmem": {
"command": "npx",
"args": ["-y", "xgmem@latest"]
}
}
Install dependencies
npm install
Build
npm run build
Run (development)
npx ts-node index.ts
Run (production)
npm start
Docker
docker build -t xgmem-mcp-server .
docker run -v $(pwd)/memories:/app/memories xgmem-mcp-server
This will persist all project memory files in the memories directory on your host.
How to Save Memory (MCP API)
To save observations (memory) for a project, call the save_project_observations tool via the MCP API:
Example JSON:
{
"name": "save_project_observations",
"args": {
"projectId": "demo-project",
"observations": [
{
"entityName": "Alice",
"contents": [
"Alice joined Acme Corp in 2021.",
"Alice is a software engineer."
]
},
{
"entityName": "Bob",
"contents": [
"Bob joined Acme Corp in 2022.",
"Bob is a product manager."
]
}
]
}
}
You can use any compatible MCP client, or send this JSON via stdin if running the server directly.
Tooling and API
xgmem exposes the following tools:
save_project_observationsget_project_observationsadd_graph_observationscreate_entitiescreate_relationsdelete_entitiesdelete_observationsdelete_relationsread_graphsearch_nodessearch_all_projectsopen_nodescopy_memory
See the get_help tool (if enabled) for documentation and usage examples via the MCP API.
Configuration
- Set
MEMORY_DIR_PATHenv variable to change the memory storage directory (default:/app/memories).
License
MIT
Recommend MCP Servers 💡
attilad/bgg-mcp-server
An MCP server integrating with BoardGameGeek XML API to enable Claude to search games, retrieve details, and access user collections/plays
yamcp
A command-line tool for organizing and managing MCP servers as local workspaces, exposing them as a unified MCP gateway for AI applications.
Higress
Higress is an AI Native API Gateway based on Istio and Envoy, which hosts MCP (Model Context Protocol) Servers through its plugin mechanism, enabling AI Agents to easily call various tools and services.
@starwind-ui/mcp
Local MCP server implementation for Starwind UI to work with AI tools like Cursor and Windsurf
hello-world
A simple FastMCP-based 'Hello World' MCP server demonstrating tools, resources, and prompts.
huggingface-mcp-server
Hugging Face MCP Server for accessing its ecosystem of Models, Datasets, Research Papers and AI tools via MCP
