pyhdb-rs
Unlock your SAP HANA data for modern analytics and AI workflows.
Pure Rust toolkit — no SAP client installation required. Connect to HANA from anywhere, stream data directly into Polars, pandas, or DuckDB via Apache Arrow. Let AI assistants explore your schemas through a secure MCP interface.
Why pyhdb-rs?
Quick Start
For Data Engineers & Scientists
pip install pyhdb_rs
from pyhdb_rs import ConnectionBuilder
import polars as pl
# Connect and extract — data flows directly to Polars without copies
conn = ConnectionBuilder.from_url("hdbsql://user:pass@hana:30015").build()
df = pl.from_arrow(conn.execute_arrow("""
SELECT material, plant, SUM(quantity) as total
FROM sapabap1.mard
GROUP BY material, plant
"""))
# Lazy evaluation for memory-efficient analytics
result = df.lazy().filter(pl.col("total") > 1000).collect()
[!TIP] Data exports via Apache Arrow — the universal columnar format. Integrates with the entire Arrow ecosystem out of the box.
Arrow Ecosystem Compatibility
Stream HANA data directly into any Arrow-compatible tool:
Zero-copy data transfer means no serialization overhead between HANA and your analytics stack.
For AI-Assisted Development
cargo install hdbconnect-mcp
Add to Claude Desktop config:
{
"mcpServers": {
"hana": {
"command": "hdbconnect-mcp",
"args": ["--url", "hdbsql://user:pass@hana:30015"]
}
}
}
Now ask Claude: "Show me the top 10 customers by revenue from VBAK/VBAP" — it queries HANA directly.
Components
Python Driver
Full DB-API 2.0 compliance with native Arrow integration. No SAP client required — works anywhere Python runs.
from pyhdb_rs.aio import ConnectionPoolBuilder
import polars as pl
pool = ConnectionPoolBuilder().url("hdbsql://user:pass@hana:30015").max_size(10).build()
async def extract_sales(region: str) -> pl.DataFrame:
async with pool.acquire() as conn:
return pl.from_arrow(await conn.execute_arrow(f"""
SELECT vbeln, erdat, netwr FROM sapabap1.vbak
WHERE region = '{region}' AND erdat >= '20240101'
"""))
# Parallel extraction across regions
results = await asyncio.gather(*[extract_sales(r) for r in ["US", "EU", "APAC"]])
import pyarrow as pa
reader = conn.execute_arrow("SELECT * FROM sapabap1.mara WHERE mtart = 'FERT'")
df = pa.RecordBatchReader.from_stream(reader).read_all().to_pandas()
from pyhdb_rs import ArrowConfig
# Process 100M rows with constant memory
config = ArrowConfig(batch_size=50_000)
reader = conn.execute_arrow("SELECT * FROM sapabap1.mseg", config=config)
for batch in reader:
process_batch(batch) # Each batch: 50K rows as Arrow RecordBatch
Full documentation: Python package README — TLS configuration, HA clusters, transaction control, error handling.
MCP Server for AI Agents
Production-ready server that exposes SAP HANA to Claude, Cline, Cursor, and any MCP-compatible assistant.
Why it matters: Instead of copy-pasting schemas or writing boilerplate queries, let AI discover and query your data directly — with guardrails.
Security by design:
- Read-only mode blocks DML/DDL by default; writes and stored procedures are opt-in, confirmed, and rate-limited
- Row limits prevent data exfiltration
- OIDC/JWT authentication for enterprise deployments
- Per-user cache isolation in multi-tenant setups
Available tools:
Full documentation: MCP server README — HTTP transport, Kubernetes deployment, Prometheus metrics.
Architecture
flowchart TB
subgraph apps["Your Applications"]
direction LR
subgraph python["Python Analytics"]
p1["ETL Pipelines"]
p2["ML Features"]
p3["BI Dashboards"]
end
subgraph ai["AI Assistants"]
m1["Claude Desktop"]
m2["Cline / Cursor"]
m3["Custom Agents"]
end
end
subgraph core["Rust Core"]
arrow["hdbconnect-arrow\nZero-copy HANA → Arrow"]
end
subgraph hana["SAP HANA"]
db[("BW/4HANA\nS/4HANA\nHANA Cloud")]
end
apps --> core
core --> hana
style apps fill:#e3f2fd
style core fill:#fff8e1
style hana fill:#e8f5e9
Requirements
- Python 3.12+ (for pyhdb_rs)
- Rust 1.96+ (for building from source or MCP server)
Resources
License
Dual-licensed under Apache-2.0 or MIT at your option.