Central-Memory-MCP is a custom fork of an MCP server, developed based on @modelcontextprotocol/server-memory, using Azure Functions MCP triggers and Table storage. It is implemented with .NET 10 (Azure Functions isolated worker) and provides persistent project memory and knowledge graph capabilities for AI assistants, with support for workspace isolation.
Purpose:
- Provide AI assistants with persistent project memory across sessions, including entities, relations, and observations.
- Support multi-developer shared workspaces, isolating the memory of different projects through workspaces.
- Provide simple HTTP tool endpoints for easy integration.
Main capabilities (current alpha stage):
- read_graph: Returns all entities in the specified workspace (relations are currently queried and connected separately).
- upsert_entity: Creates or updates an entity (if the name exists, the existing Id is retained).
- upsert_relation: Creates or updates a relation between two entities (requires a GUID or a resolvable name).
- get_entity_relations: Lists the relations emitted by a specific entity.
- Health and readiness endpoints: /api/health and /api/ready.
- Planned but not yet implemented features: search_entities, search_relations, stats, temporal, batch operations, merge/detect duplicates, etc.
Usage:
- To read the graph, call the read_graph MCP tool from the client and provide the workspaceName parameter.
- Ensure consistent use of workspaceName; workspaceId in the documentation has been replaced with workspaceName.
- upsert preserves entity identity by looking up (WorkspaceName + Name).
- upsert_relation requires an entity GUID or a resolved name; it fails if the name is missing.
Data model:
- Entities table: PartitionKey = WorkspaceName, RowKey = Guid (Id).
- Relations table: PartitionKey = WorkspaceName, RowKey = Guid (Id).
- Workspaces table (future extension).
- Observations are internally stored as a "||"-separated string and split into a list when read.
Applicable scenarios:
- AI assistant projects that need memory across sessions.
- Multi-developer collaborative projects that need a shared knowledge graph.
- A lightweight memory service based on Azure Functions and Table storage.
Limitations:
- Currently in the alpha stage, some features (search, pagination, statistics, batch operations, etc.) are not yet implemented.
- Relation queries and connections may not be efficient.
- Requires an Azure Functions and Table storage environment.
- Logging and telemetry (Application Insights connection) are future enhancements.
