Memgraph AI Toolkit

Build powerful AI applications with graph-powered RAG using Memgraph. This toolkit provides everything you need to integrate knowledge graphs into your GenAI workflows.

🚀 Quick Setup

Start Memgraph

docker run -p 7687:7687 \
  --name memgraph \
  memgraph/memgraph-mage:latest \
  --schema-info-enabled=true

Install Packages

# Core toolbox
pip install memgraph-toolbox

# LangChain integration
pip install langchain-memgraph

# MCP server
pip install mcp-memgraph

# Unstructured to Graph
pip install unstructured2graph

📚 Usage Examples

Context Graph - Capture Your Agent Sessions

Turn your coding agents' sessions (Claude Code, Codex, Copilot CLI, Cursor, OpenCode, Antigravity CLI, Grok Build) into a queryable knowledge graph. Every session records the tools it called, the skills it used, and the memories it wrote — all joined on a shared (:Session) node in Memgraph.

One command finds or starts Memgraph, wires every agent it finds, sets your identity and checks the result (interactive on a terminal):

curl -fsSL https://raw.githubusercontent.com/memgraph/ai-toolkit/main/context-graph/scripts/install.sh | bash

Query across every session — e.g. which skills a user has used:

MATCH (:User {user_id: "your-name"})-[:HAD_SESSION]->(:Session)-[:USED_SKILL]->(s:Skill)
RETURN s.name, count(*) AS uses ORDER BY uses DESC;

👉 Context Graph guide — components, Codex setup, SDK usage, and reconciling sessions into an entity graph.

unstructured2graph - Build Knowledge Graphs from Documents

Transform PDFs, URLs, and documents into queryable knowledge graphs:

import asyncio
from memgraph_toolbox.api.memgraph import Memgraph
from lightrag_memgraph import MemgraphLightRAGWrapper
from unstructured2graph import LightRAGBackend, from_unstructured


async def main():
    memgraph = Memgraph()

    lightrag = MemgraphLightRAGWrapper()
    await lightrag.initialize(working_dir="./lightrag_storage")

    # Ingest documents from URLs or local files
    await from_unstructured(
        sources=["https://example.com/doc.pdf", "./local_file.md"],
        memgraph=memgraph,
        extraction_backend=LightRAGBackend(lightrag),
        link_chunks=True,
        enforce_ontology=True,  # promote entity_type to real labels (:Person, :Organization, ...)
    )
    await lightrag.afinalize()


asyncio.run(main())

extraction_backend is pluggable — LightRAGBackend (LLM-based, shown above) or GLiNER2Backend (local, LLM-free) — see unstructured2graph § choosing an extraction backend.

👉 Full Documentation | Examples

langchain-memgraph - LangChain Integration

Natural Language Queries with MemgraphQAChain

from langchain_memgraph.graphs.memgraph import MemgraphLangChain
from langchain_memgraph.chains.graph_qa import MemgraphQAChain
from langchain_openai import ChatOpenAI

graph = MemgraphLangChain(url="bolt://localhost:7687")

chain = MemgraphQAChain.from_llm(
    ChatOpenAI(temperature=0),
    graph=graph,
    model_name="gpt-4-turbo",
    allow_dangerous_requests=True,
)

response = chain.invoke("Who are the main characters in the dataset?")
print(response["result"])

Build Agents with MemgraphToolkit

from langchain.chat_models import init_chat_model
from langchain_memgraph import MemgraphToolkit
from langchain_memgraph.graphs.memgraph import MemgraphLangChain
from langgraph.prebuilt import create_react_agent

llm = init_chat_model("gpt-4o-mini", model_provider="openai")
db = MemgraphLangChain(url="bolt://localhost:7687")
toolkit = MemgraphToolkit(db=db, llm=llm)

agent = create_react_agent(llm, toolkit.get_tools())
events = agent.stream({"messages": [("user", "Find all Person nodes")]})

👉 Full Documentation

mcp-memgraph - Model Context Protocol Server

Expose Memgraph to LLMs via MCP. Run with Docker:

# HTTP mode (recommended)
docker run --rm -p 8000:8000 memgraph/mcp-memgraph:latest

# Stdio mode for MCP clients
docker run --rm -i -e MCP_TRANSPORT=stdio memgraph/mcp-memgraph:latest

Available Tools:

👉 Full Documentation

sql2graph Agent - Automated Database Migration

Migrate from MySQL/PostgreSQL to Memgraph with AI assistance:

cd agents/sql2graph
uv run main.py

👉 Full Documentation

🛠️ Packages Overview

Context Graph — capture agent sessions

A family of components that persist your coding-agent sessions (Claude Code, Codex, Copilot CLI, Cursor, OpenCode, Antigravity CLI, Grok Build) into one Memgraph graph. See the Context Graph guide.

❓ FAQ

Which databases are supported? Memgraph is the primary target. The sql2graph agent supports MySQL and PostgreSQL as source databases.

Do I need an LLM API key? Yes, for features like entity extraction (unstructured2graph) and natural language queries (langchain-memgraph).

Can I use local LLMs? Yes! LangChain integration supports any LangChain-compatible model, including Ollama.

🤝 Community

• GitHub Issues
• Discord
• Documentation

⭐ If you find this toolkit helpful, please star the repository!

🧪 Developing Locally

You can build and test each package directly from your repo.

Core tests

uv pip install -e memgraph-toolbox[test]
pytest -s memgraph-toolbox/src/memgraph_toolbox/tests

LangChain integration tests

Create a .env file with your OPENAI_API_KEY, as the tests depend on LLM calls:

uv pip install -e integrations/langchain-memgraph[test]
pytest -s integrations/langchain-memgraph/tests

MCP integration tests

uv pip install -e integrations/mcp-memgraph[test]
pytest -s integrations/mcp-memgraph/tests

Context Graph tests

The Context Graph components (and unstructured2graph) test against a live Memgraph. scripts/dev-memgraph.sh owns that lifecycle — it starts an isolated instance, runs each component's suite against it, and tears down:

./scripts/dev-memgraph.sh up
./scripts/dev-memgraph.sh test          # all components; or e.g. `test sessions-graph`
./scripts/dev-memgraph.sh down

sql2graph agent

To run a complete database migration workflow with the agent:

cd agents/sql2graph
uv run main.py

Note: The agent requires both MySQL and Memgraph connections. Set up your environment variables in .env based on .env.example.

If you are running any test on macOS in zsh, add "" to the command:

uv pip install -e memgraph-toolbox"[test]"