Memgraph AI Toolkit MCP server

Memgraph AI Toolkit

Python packages for building AI applications with Memgraph: agent memory, knowledge graphs, GraphRAG, and framework integrations.

Use the packages independently, or combine them to bring documents, databases, and agent activity into a graph.

Documentation · Examples · Discord · Issues

Choose your starting point

I want to…	Start here
Give coding agents memory across sessions and tools	Context Graph
Extract entities and relationships from documents	unstructured2graph
Query Memgraph in natural language or build a LangChain agent	langchain-memgraph
Connect an MCP client to Memgraph	mcp-memgraph
Use Memgraph as LightRAG storage	lightrag-memgraph
Migrate a MySQL or PostgreSQL database into a graph	sql2graph
Work directly with Memgraph from Python	memgraph-toolbox
Define and validate a graph's entity and relation types	hygm

Agent memory with Context Graph

Keep what your agents learn in a database you control. Context Graph captures sessions, tool calls, and skill usage, stores explicit memory, and retrieves relevant past conversations for the next session.

It supports Claude Code, Codex, Copilot CLI, Cursor, OpenCode, Antigravity CLI, and Grok Build, plus applications built with the Claude Agent SDK or OpenAI Agents SDK.

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

Setup installs the CLI and connectors, starts its own Memgraph container or connects to one you specify, and wires the agents you select. Docker is needed if setup starts Memgraph for you.

Get started with Context Graph · Components · Recall benchmarks

Knowledge graphs from documents

unstructured2graph parses documents and extracts entities and relationships. Its default backend, GLiNER2, runs locally without an LLM API key; LightRAG is available as an LLM-based alternative.

Start Memgraph:

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

Install the package:

pip install unstructured2graph

Extract a graph from text:

import asyncio
from memgraph_toolbox.api.memgraph import Memgraph
from unstructured2graph import from_texts
from unstructured2graph.gliner2_backend import GLiNER2Backend


async def main():
    await from_texts(
        texts=["Alice works for Acme in London."],
        memgraph=Memgraph(),
        extraction_backend=GLiNER2Backend(),
        enforce_ontology=True,
    )


asyncio.run(main())

The extraction model downloads on first use. See unstructured2graph for files, URLs, document-format extras, and custom ontologies.

Framework integrations

Install only the integration you need:

pip install langchain-memgraph  # graph store, QA chain, and agent toolkit
pip install mcp-memgraph       # MCP server for database queries and schema tools
pip install lightrag-memgraph  # LightRAG storage backends

Each package's README covers its connection settings and examples. LLM-based features need credentials for the model provider you choose; local extraction, capture, and recall do not require an LLM key.

Developing locally

This is a Python 3.10+ uv workspace. From the repository root:

uv sync --all-packages --all-extras
uvx ruff@0.16.0 check .
uvx ruff@0.16.0 format --check .
uvx ty@0.0.75 check .

These are the versions CI pins in the lint workflow.

Context Graph, HyGM, and unstructured2graph use an isolated test Memgraph managed by the development script:

./scripts/dev-memgraph.sh test
./scripts/dev-memgraph.sh test sessions-graph  # one package
./scripts/dev-memgraph.sh test-down           # remove the test container

For other packages, see their READMEs and the test workflow for dependencies and credentials. The development script also manages a separate exploration database for dogfooding.

Community

Ask questions on Discord, report bugs or request features in GitHub Issues, and explore the Memgraph AI documentation.

MIT license