memgraph-sdk 0.3.0

Official Python SDK for Memgraph - the memory graph for AI agents

Memgraph SDK

The official Python SDK for Memgraph -- the memory graph for AI agents.

Installation

pip install memgraph-sdk

For async support:

pip install "memgraph-sdk[async]"

Quick Start

from memgraph_sdk import MemgraphClient

client = MemgraphClient(
    api_key="mg_your_api_key",
    tenant_id="your-tenant-id",
)

# Store a memory
client.add(
    text="User prefers dark mode and uses PyTorch.",
    user_id="user_123",
)

# Search for relevant context
result = client.search(
    query="What does this user prefer?",
    user_id="user_123",
)
print(result)

Async Client

from memgraph_sdk import AsyncMemgraphClient

async with AsyncMemgraphClient(
    api_key="mg_your_api_key",
    tenant_id="your-tenant-id",
) as client:
    await client.add("User prefers dark mode", user_id="user_123")
    result = await client.search("user preferences", user_id="user_123")

Memory Intelligence API

# Memory health metrics
health = client.health(user_id="user_123")

# Detect contradictions
contradictions = client.contradictions(user_id="user_123")

# Evaluate retrieval quality
evaluation = client.evaluate(query="test query", user_id="user_123")

# Cognitive Integrity Score (0-100)
score = client.mcis(user_id="user_123")

# Run benchmarks
result = client.benchmark(scenario="contradiction_storm")
scenarios = client.benchmark_scenarios()

CLI

# Initialize Memgraph in your project
memgraph init

# Store a memory
memgraph remember "We decided to use PostgreSQL" -c decision

# Search memories
memgraph recall "database choice"

# Check connection
memgraph status

Configuration

The client reads the API URL from the MEMGRAPH_API_URL environment variable, defaulting to http://localhost:8001/v1. You can also pass it explicitly:

client = MemgraphClient(
    api_key="mg_your_key",
    tenant_id="your-tenant-id",
    base_url="https://api.memgraph.ai/v1",
)

Examples

See the examples/ directory for complete integration examples:

• Quick Start -- Basic add and search
• Agent Integration -- OpenAI-powered agent with memory
• MCP Server -- Model Context Protocol server for Claude/Cursor
• LangChain -- LangChain integration
• OpenAI -- OpenAI integration
• Benchmarks -- Performance benchmarking

License

MIT License. See LICENSE for details.