MCP Qdrant Memory – A Knowledge Graph Server for AI Assistants

The MCP Qdrant Memory server addresses a common pain point in building intelligent AI assistants: persisting and retrieving structured knowledge while also enabling semantic search. Traditional key‑value stores or simple in‑memory caches fall short when an assistant must understand relationships between concepts and surface contextually relevant information. This server solves that by combining a graph‑based knowledge representation with a vector database, giving developers a single endpoint to create entities, define relationships, and perform similarity searches in one cohesive workflow.

At its core, the server stores a full knowledge graph in a lightweight JSON file (). This file contains entities, relations, and observations that can be queried quickly for graph‑centric operations such as adding or deleting nodes. For semantic search, the server automatically generates embeddings using OpenAI’s models and pushes those vectors to a Qdrant collection. The dual‑storage design keeps the graph fast for direct lookups while leveraging Qdrant’s efficient vector similarity engine to surface concepts that match a user’s intent, even when the wording differs.

Key capabilities include:

• Entity and Relation Management: Tools to create, delete, and read entities and relations, allowing developers to build rich ontologies tailored to their domain.
• Observations: Arbitrary text snippets attached to entities, providing contextual details that enrich embeddings.
• Semantic Search (): Accepts a free‑text query, generates an embedding, and retrieves the most similar entities or relations from Qdrant. Results are automatically ranked by similarity score.
• Persistent Synchronization: Every mutation updates the JSON graph and, in parallel, regenerates embeddings for Qdrant, ensuring consistency between the two storage layers.
• HTTPS and Docker Support: Ready for production deployment behind reverse proxies or as a containerized microservice.

Typical use cases include:

• Conversational Agents: An assistant can answer “What are the main tasks in Project X?” by querying the graph, then surface related concepts like “deadline” or “resource allocation” through semantic search.
• Domain Knowledge Bases: Companies can model product catalogs, technical documentation, or policy documents as a graph and let the assistant retrieve related items even when users phrase queries differently.
• Dynamic Knowledge Updates: Teams can incrementally add new entities or observations without re‑building the entire index, enabling continuous learning.

Integrating this server into an MCP workflow is straightforward: developers add the server to their MCP settings, expose the defined tools (, , etc.), and then invoke those tools from their AI assistant’s prompt logic. Because the server handles both structured graph queries and vector search, developers can design assistants that combine deterministic knowledge retrieval with flexible semantic matching—all through the same MCP interface.