MemoryMesh

MemoryMesh is an MCP server that provides a local knowledge graph memory system for AI models, with a primary focus on text-based RPGs, interactive storytelling, and structured simulations. It connects local AI clients such as Claude Desktop, Cursor, and VS Code to an extensible graph database stored locally in JSON format. Game masters, worldbuilders, and AI developers use MemoryMesh to maintain persistent, consistent entities and relationships across ongoing conversational sessions. The server dynamically analyzes user-defined JSON schema blueprints placed in its data directory and automatically exposes custom tools for adding, updating, and deleting specific node types and relational edges. By translating structured schemas into model tools, it enforces required fields, validates enumerated attributes, and ensures models create directional edges between linked entities. MemoryMesh also pairs with companion tools like the visual SchemaManager and Memory Viewer to let users inspect and modify graph data outside the AI context window.

Category: AI Memory & Context

Tags: knowledge-graph, memory, rpg, storytelling

Visit MemoryMesh

How to install and configure MemoryMesh

1. Clone or download the MemoryMesh repository from https://github.com/CheMiguel23/MemoryMesh. 2. In the project directory, run npm install to install dependencies and compile the server. 3. Ensure your custom or default .schema.json files are placed in the dist/data/schemas directory. 4. Open your MCP client configuration file (such as your Claude Desktop or Cursor configuration). 5. Register the MemoryMesh server under your MCP server settings pointing to the built Node.js executable. Consult the repository README for specific command parameters and exact JSON configuration structures: https://github.com/CheMiguel23/MemoryMesh.

What you can do with MemoryMesh

• Managing character attributes, races, and inventory edges across prolonged interactive text-based role-playing campaigns. - Enforcing worldbuilding rules and spatial locations by automatically linking non-player characters to specific graph locations. - Simulating social relationships and organizational structures through directed, weighted graph edges between interacting entities. - Generating custom CRUD toolsets dynamically simply by placing JSON schema definitions into the server schema folder. - Inspecting entity connections, node properties, and graph statistics externally using the standalone web-based Memory Viewer.

Key facts

• https://github.com/CheMiguel23/MemoryMesh
• AI Memory & Context, Notes & Knowledge Management
• knowledge-graph, memory, rpg, storytelling

Part of MCP Servers

What can MemoryMesh do?

MemoryMesh acts as a structured graph memory server for AI models. It reads JSON schemas to automatically generate dynamic tools for creating, updating, and deleting nodes and edges. It persists entity data and relational links to a local memory file, allowing language models to preserve consistent world states, characters, and events across multiple conversational interactions.

Which MCP clients work with MemoryMesh?

MemoryMesh supports clients compliant with the Model Context Protocol specification. Because its SDK was updated to version 1.25.2, it is compatible with modern MCP hosts including Claude Desktop, Cursor, ChatGPT, Gemini, and VS Code.

Where does MemoryMesh store its knowledge graph data?

By default, MemoryMesh saves its knowledge graph data in a JSON file located at dist/data/memory.json within the project folder. Schemas defining entity structures and dynamic tools are located in the dist/data/schemas directory.

Is MemoryMesh open source?

Yes, MemoryMesh is open source software published under the MIT license on GitHub.

How does MemoryMesh generate dynamic tools?

MemoryMesh automatically inspects JSON schema files whose names begin with add_ located in its schemas directory. Based on the properties, requirements, and relationships declared in each schema, the server generates corresponding add, update, and delete tools for the AI model to execute.