petro-mcp · Free & Open Source · Alpha v0.8.1
70 tested tools for LAS parsing, decline curves, PVT, drilling, and economics. Your team asks questions in plain English; petro-mcp runs deterministic code underneath — not AI-generated math.
Large language models are powerful reasoners but have no native way to open an LAS file, fit an Arps decline, or compute a kill mud weight. petro-mcp closes that gap.
Install via pip, Glama, or source. About five minutes end-to-end for a first-time setup.
Recommended
pip install petro-mcp
Then add petro-mcp to your Claude Desktop, Cursor, or VS Code MCP config. Full setup instructions are in the README.
Package Index
Standard Python package install via pip or uv.
View on PyPI →
Verified Listing
Curated MCP marketplace with build tests, schema inspection, and a Docker image.
View on Glama →
Source
Full source, issue tracker, example notebooks, and contribution guide.
View on GitHub →
Also listed on mcp.so, the community MCP directory.
70 tools across 12 petroleum engineering domains. Load only the groups you need to keep context overhead low.
LAS 2.0 parsing, header extraction, curve inspection, depth-range filtering. Built on lasio.
Query production CSVs by well and date range. Detect shut-ins, rate jumps, water breakthrough, GOR blowouts.
Arps (exp, hyp, harmonic), modified hyperbolic, Duong, PLE, SEPD, THM models. EUR calculation with economic limits.
Bubble point, solution GOR, oil FVF, viscosity, gas compressibility — standard industry correlations.
Shale volume, effective porosity, water saturation (Archie, Simandoux), net pay cutoffs from log curves.
Hydraulics, ECD, kill mud weight, kick tolerance, pressure loss across the circulation system.
Material balance, radial flow, skin, pseudo-steady-state productivity index calculations.
IPR and TPR curves, nodal analysis, tubing sizing, artificial lift screening.
NPV, IRR, payout, break-even oil price. Configurable price decks, severance tax, and working interest.
Minimum-curvature survey math, dogleg severity, 3D wellpath generation (optional, via welleng).
Field-standard unit conversions and quick-look formulas that engineers reach for every day.
These are the prompts engineers actually use. Your AI assistant calls the right petro-mcp tools under the hood.
“Read the LAS file at /data/wells/wolfcamp.las and summarize the curves.”
“Load production data from /data/prod.csv and fit a hyperbolic decline for Wolfcamp A-1H.”
“Calculate EUR with qi=800, Di=0.06, b=1.1 and a 5 bbl/day economic limit.”
“Run a nodal analysis: 3,500 psi reservoir, PI = 4.2, 2.875-inch tubing, 150 psi wellhead.”
“What’s the kill mud weight if SIDP is 500 psi, current MW is 10.5 ppg, TVD 12,000 ft?”
“Calculate PVT properties for 35° API oil at 200 °F and 3,000 psi.”
MCP is an open standard. petro-mcp works with any compliant client.
Drop into claude_desktop_config.json.
Add to .cursor/mcp.json.
MCP extension, same JSON config.
Any MCP-compatible runtime — the protocol is open.
Skip the scripting loop. Ask your AI assistant for the same LAS parse, decline fit, or PVT calc you already run — and get an answer backed by tested code, not hallucinated math.
A drop-in MCP layer over the petroleum workflows your engineers already trust. Pair with your own tools or extend the codebase — MIT license, no lock-in.
Teach, prototype, and publish with a shared, inspectable foundation. Every calculation is transparent source code — not a black-box API.
petro-mcp is built and maintained by Groundwork Analytics, an AI and data science firm working with upstream operators across the Permian, Eagle Ford, and other major U.S. basins. Led by Dr. Mehrdad Shirangi, Stanford PhD in energy systems optimization.
petro-mcp runs as a local process — your LAS files, production CSVs, and well data never leave your machine. What crosses the network is the prompt you type and the tool outputs your AI assistant sees, which go to your chosen LLM provider (Claude, OpenAI, etc.). For full data isolation, pair petro-mcp with a self-hosted model (Ollama, vLLM, or your own endpoint) — the protocol is provider-agnostic.
About five minutes to install and configure. Point it at an LAS or production CSV and ask a question.
The open-source foundation
petro-mcp is free, MIT-licensed, and stays that way.
Need to scale this across your fleet — private-cloud hosting, ingestion into your wells database, governance for shared teams, or custom tools on top? That’s Groundwork’s paid engagement, not petro-mcp itself.
Book a Discovery Call →