pyTigerGraph

pyTigerGraph is a Python client for TigerGraph databases. It wraps the REST++ and GSQL APIs and provides both a synchronous and an asynchronous interface.

Full documentation: https://docs.tigergraph.com/pytigergraph/current/intro/

Downloads: | |

---

Installation

Base package

pip install pyTigerGraph

Optional extras

Extra	What it adds	Install command
gds	Graph Data Science — data loaders for PyTorch Geometric, DGL, and Pandas	pip install 'pyTigerGraph[gds]'
mcp	Model Context Protocol server — installs pyTigerGraph-mcp (convenience alias)	pip install 'pyTigerGraph[mcp]'
fast	orjson JSON backend — 2–10× faster parsing, releases the GIL under concurrent load	pip install 'pyTigerGraph[fast]'

Extras can be combined:

pip install 'pyTigerGraph[fast,gds,mcp]'

[gds] prerequisites

Install torch before installing the gds extra:

1. Install Torch
2. Optionally Install PyTorch Geometric or Install DGL
3. pip install 'pyTigerGraph[gds]'

[fast] — orjson JSON backend

orjson is a Rust-backed JSON library that is detected and used automatically when installed. No code changes are required. It improves throughput in two ways:

• Faster parsing — 2–10× vs stdlib json
• GIL release — threads parse responses concurrently instead of serialising on the GIL

If orjson is not installed the library falls back to stdlib json transparently.

---

Quickstart

Synchronous connection

from pyTigerGraph import TigerGraphConnection

conn = TigerGraphConnection(
    host="http://localhost",
    graphname="my_graph",
    username="tigergraph",
    password="tigergraph",
)

print(conn.echo())

Use as a context manager to ensure the underlying HTTP session is closed:

with TigerGraphConnection(host="http://localhost", graphname="my_graph") as conn:
    result = conn.runInstalledQuery("my_query", {"param": "value"})

Asynchronous connection

AsyncTigerGraphConnection exposes the same API as TigerGraphConnection but with async/await syntax. It uses aiohttp internally and shares a single connection pool across all concurrent tasks, making it significantly more efficient than threaded sync code at high concurrency.

import asyncio
from pyTigerGraph import AsyncTigerGraphConnection

async def main():
    async with AsyncTigerGraphConnection(
        host="http://localhost",
        graphname="my_graph",
        username="tigergraph",
        password="tigergraph",
    ) as conn:
        result = await conn.runInstalledQuery("my_query", {"param": "value"})
        print(result)

asyncio.run(main())

Token-based authentication

conn = TigerGraphConnection(
    host="http://localhost",
    graphname="my_graph",
    gsqlSecret="my_secret",   # generates a session token automatically
)

HTTPS / TigerGraph Cloud

conn = TigerGraphConnection(
    host="https://my-instance.i.tgcloud.io",
    graphname="my_graph",
    username="tigergraph",
    password="tigergraph",
    tgCloud=True,
)

---

Connection parameters

Parameter	Type	Default	Description
host	str	"http://127.0.0.1"	Server URL including scheme (http:// or https://)
graphname	str	""	Target graph name
username	str	"tigergraph"	Database username
password	str	"tigergraph"	Database password
gsqlSecret	str	""	GSQL secret for token-based auth (preferred over username/password)
apiToken	str	""	Pre-obtained REST++ API token
jwtToken	str	""	JWT token for customer-managed authentication
restppPort	int|str	"9000"	REST++ port (auto-fails over to 14240/restpp for TigerGraph 4.x)
gsPort	int|str	"14240"	GSQL server port
certPath	str	None	Path to CA certificate for HTTPS
tgCloud	bool	False	Set to True for TigerGraph Cloud instances

---

Performance notes

Synchronous mode (TigerGraphConnection)

• Each thread gets its own dedicated HTTP session and connection pool, so concurrent threads never block each other.
• Install pyTigerGraph[fast] to activate the orjson backend and reduce JSON parsing overhead under concurrent load.
• Use ThreadPoolExecutor to run queries in parallel:

from concurrent.futures import ThreadPoolExecutor, as_completed

with TigerGraphConnection(...) as conn:
    with ThreadPoolExecutor(max_workers=16) as executor:
        futures = [executor.submit(conn.runInstalledQuery, "q", {"p": v}) for v in values]
        for f in as_completed(futures):
            print(f.result())

Asynchronous mode (AsyncTigerGraphConnection)

• Uses a single aiohttp.ClientSession with an unbounded connection pool shared across all concurrent coroutines — no GIL, no thread-scheduling overhead.
• Typically achieves higher QPS and lower tail latency than the threaded sync mode for I/O-bound workloads.

import asyncio
from pyTigerGraph import AsyncTigerGraphConnection

async def main():
    async with AsyncTigerGraphConnection(...) as conn:
        tasks = [conn.runInstalledQuery("q", {"p": v}) for v in values]
        results = await asyncio.gather(*tasks)

asyncio.run(main())

---

Graph Data Science (GDS)

The gds sub-module provides data loaders that stream vertex and edge data from TigerGraph directly into PyTorch Geometric, DGL, or Pandas DataFrames for machine learning workflows.

Install requirements, then access via conn.gds:

conn = TigerGraphConnection(host="...", graphname="...")
loader = conn.gds.vertexLoader(attributes=["feat", "label"], batch_size=1024)
for batch in loader:
    train(batch)

See the GDS documentation for full details.

---

MCP Server

The TigerGraph MCP server is now a standalone package: pyTigerGraph-mcp. It exposes TigerGraph operations as tools for AI agents and LLM applications (Claude Desktop, Cursor, Copilot, etc.).

# Recommended — install the standalone package directly
pip install pyTigerGraph-mcp

# Or via the pyTigerGraph convenience alias (installs pyTigerGraph-mcp automatically)
pip install 'pyTigerGraph[mcp]'

# Start the server (reads connection config from environment variables)
tigergraph-mcp

For full setup instructions, available tools, configuration examples, and multi-profile support, see the pyTigerGraph-mcp README.

Migrating from pyTigerGraph.mcp? Update your imports:

# Old
from pyTigerGraph.mcp import serve, ConnectionManager
# New
from tigergraph_mcp import serve, ConnectionManager

---

Getting started video

Companion notebook: Google Colab

---

Links

• Documentation
• PyPI
• GitHub Issues
• Source