Heliciel — Propeller & Wing Design

A real session — a 1140 mm, 3-blade marine propeller at 105 rpm: 1002.7 N of thrust, 222.4 N·m of torque, 2.45 kW, 92 % efficiency. Every figure computed by Heliciel, none of it written by the model. The viewer renders inline in ChatGPT; in Claude the same viewer opens from a link.

Ask any assistant to "design a propeller" and it will produce something plausible: a diameter, a blade count, an efficiency figure. None of it is computed. For a conversation piece that is fine; for something you are going to machine, print or bolt onto a boat, it is worse than useless — a wrong number that looks right.

This server hands the assistant a real solver. It refuses to invent: missing parameters are asked for, values inherited from a loaded model are flagged as inherited, and a failed computation is reported as-is rather than replaced by an estimate.

Aircraft propellers · drone and multirotor propellers · marine propellers · axial fans · wind turbines · hydro turbines, including ducted Kaplan runners · wings and hydrofoils.

"I have an 8 kg drone with 4 motors and a 150% thrust margin. Size the propellers."

"Design an aircraft propeller: 50 kW, 120 km/h, max diameter 1.5 m, 2 blades."

"Load a wind turbine, 3 blades, 4 m diameter, 8 m/s wind — what power can I harvest?"

"Design a Kaplan runner for 5 m³/s and a 4 m net head, then show me the efficiency curve."

"Compare 2 vs 3 blades and show me the best one in 3D."

Each pass getsits own private Heliciel instance: the real Windows application, driven through its own interface rather than through a reimplementation of it. No state is shared between users. Your design projects are not archived — the session folder is erased when the instance is recycled, and download links expire after 60 minutes.

That architecture is why the numbers are the software's own. It is also why the first call of a session can take up to two and a half minutes if an instance has to start, and why long optimizations return progress milestones rather than blocking: past 120 s a tool answerscommande_en_cours, andlire_etat_commandereports the generations as they complete.

- Every number given to the user comes from a tool result. The server instructs the model never to invent or estimate one.
- Images come from the software's own screen. The model is instructed never to generate a picture of Heliciel or of its results — a generated image of engineering results is a fabrication, whatever it looks like.
- Values inherited from the loaded model, rather than stated by the user, are returned flagged as inherited.
- A quantity that has not been computed isnot published. It is returned as unavailable, with the reason and the tool to call — never as a zero that would read as a result.
- Heliciel's own design alerts travel with the results, so a computation that succeeds while flagging a geometry to rework says so.

- Performance isideal and theoretical: surfaces are treated as aerodynamically smooth, profile lift and drag data are generated by XFOIL, and fluid compressibility is not taken into account — precision therefore drops at transonic and supersonic speeds.
- No acoustic model: blade-passing frequencies and tip speed are computed, an SPL dB(A) level at a distance is not — and is not estimated in its place.
- Off-design and reverse-engineering modes are less precise than a computation at optimal twist.
- This is design assistance. Test your build, and do not put anyone at risk on the strength of a computation alone.

The Heliciel software itself is proprietary and is not published here. This repository holds the public description of the MCP service.

Published byMecaflux / Heliciel— fluid-mechanics software since 2005.

A lightweight MCP (Model Context Protocol) server for Blender. It offers a natural language interface with Blender’s Python API, improving access to documentation, and allowing users to explore and understand complex setups.

The Lottie Creator MCP brings your AI assistant directly into your animation workflow — giving it full access to LottieFiles Creator so it can build and edit Lottie animations on your behalf through natural language.

AI Video, Image & Audio Generation with over 150 models

Rive MCP let AI handle repetitive tasks, like creating complex View Models, State Machines with hundreds of states/layers, Layouts, Shapes, and more.

Create, edit, and export SVGator animations (SVG, Lottie, GIF, video) from your AI assistant, and keep an editable project you can open in the editor anytime.

You're agent can Chain 60+ AI image and video models on one workflow canvas

Real interaction references and motion recipes for AI coding agents.

Search 1,145 real website designs by style, font, colour, section and measured design tokens.

Render, verify, describe, and safely edit Mermaid diagrams through MCP.