Memgraph AI Toolkit: A Codebase for Graph-Powered Applications
Updated October 2026: The AI Toolkit has grown since we published this post. It now includes separate Python packages for AI memory, creating knowledge graphs from documents and SQL databases, GraphRAG, and agent and framework integrations.
That includes Context Graph, Memgraph’s AI memory for keeping context across agent sessions and tools. See the AI Toolkit docs for the current package list, or AI Memory on Memgraph to get started in one command.
Introduction
With the introduction of structured outputs and function calling in large language models (LLMs), developers can now integrate dynamic function execution directly into their application pipelines. LLMs can invoke functions to fetch real-time data or output structured JSON, which can then be passed along to the next model call.
These capabilities support high-level frameworks like LangGraph, LlamaIndex, and the Model Context Protocol (MCP), enabling the development of sophisticated, agent-based workflows.
Within this context, a function is typically wrapped as a tool, a stateless and reusable component that an agent (LLM) can invoke to perform a specific task.
For example, a run_cypher_query() tool allows an agent to execute Cypher queries when interacting with a graph database like Memgraph.
To streamline the development of these tools across multiple frameworks, we’ve introduced the Memgraph Toolbox, a central part of the Memgraph AI Toolkit.
Memgraph Toolbox
LLM-Powered Agents can autonomously decide which tools to use based on the prompt context or user queries. Given Memgraph’s capabilities as a graph database, agents have many ways to interact with the graph, from exploring schema and running queries to calculating graph algorithms.
To avoid duplicating tool implementations across different frameworks and repositories, the Memgraph Toolbox offers a shared collection of Python-based tools that adhere to the DRY principle. This toolbox can be reused across LangGraph, LlamaIndex, MCP, and other agentic environments.
Installation
To install the toolbox:
pip install memgraph-toolboxExample Usage
For usage examples and the current list of tools, see the toolbox README. To see how agents call these tools from LangGraph or MCP, see the docs.
Moving MCP and LangGraph into the AI Toolkit
If you missed our previous releases, here’s the background:
We have moved the Memgraph LangChain integration and the MCP server into a single monorepo, the AI Toolkit. Each is its own package there, langchain-memgraph and mcp-memgraph, and installs on its own.
Contribute or Build With Us
The Memgraph AI Toolkit is open-source and lives under the Memgraph GitHub organization:
🔗 https://github.com/memgraph/ai-toolkit
If you’re building LangGraph workflows, graph-native RAG pipelines, AI memory, or agent-based systems using Memgraph, we’d love your feedback. Contributions and issues are always welcome. Join Memgraph community on Discord to connect with other developers!
