.codex-plugin/plugin.json

{
  "author": {
    "name": "JEAN FRANCOIS IGLESIAS DE DIOS"
  },
  "description": "Heliciel turns your AI assistant into a propeller and turbine design engineer. Instead of plausible-looking guesses, every number comes from a real BEM (Blade Element Momentum) computation run by the Heliciel engineering software \u2014 the same solver used by hobbyists and professionals for 15+ years.\n\nWhat you can design: aircraft propellers, drone/multirotor propellers, marine propellers, axial fans, wind turbines, hydro turbines (including ducted Kaplan runners), wings and hydrofoils.\n\nWhat the AI can do with it:\n- Size a propeller from your specification (power, speed, thrust, diameter, blade count) \u2014 the design RPM is searched automatically to hit your target.\n- Analyze off-design behaviour: operating-point performance, sweep curves (power vs wind speed, efficiency vs RPM), 2D performance maps.\n- Optimize: RPM, blade count, tip radius, chord distribution \u2014 iterative or genetic search, always within bounds you choose.\n- Check physics: cavitation (marine), noise indicators, blade stress warnings, hydraulic head balance for ducted turbines.\n- Show results: interactive 3D viewer embedded in the conversation (rotate the blade, pick an element, read its profile), performance charts, design alerts.\n- Export for manufacturing: STL (3D printing/CFD), IGES (CAD), OBJ, blade sections (JSON), analysis tables (CSV).\n\nHonest by design: the server refuses to invent values. Missing parameters are asked, not assumed; inherited model values are flagged; failed computations are reported as-is with Heliciel's own design alerts.\n\nRequirements: an existing Heliciel MCP pass. Sign in with OAuth: your pass key is your credential \u2014 no account, no password to create.",
  "interface": {
    "capabilities": [],
    "category": "Education & Research",
    "defaultPrompt": [
      "Load a 4 m wind turbine, 3 blades, in 8 m/s wind: give me Cp and power, then optimize the rotation speed.",
      "Design an aircraft propeller: 50 kW at 33.3333 m/s (120 km/h), 1.5 m diameter, 2 blades.",
      "Design a hovering drone propeller: 400 mm diameter at 5000 rpm. What thrust does each one give?"
    ],
    "developerName": "JEAN FRANCOIS IGLESIAS DE DIOS",
    "displayName": "Heliciel Propeller Design",
    "longDescription": "Heliciel turns your AI assistant into a propeller and turbine design engineer. Instead of plausible-looking guesses, every number comes from a real BEM (Blade Element Momentum) computation run by the Heliciel engineering software \u2014 the same solver used by hobbyists and professionals for 15+ years.\n\nWhat you can design: aircraft propellers, drone/multirotor propellers, marine propellers, axial fans, wind turbines, hydro turbines (including ducted Kaplan runners), wings and hydrofoils.\n\nWhat the AI can do with it:\n- Size a propeller from your specification (power, speed, thrust, diameter, blade count) \u2014 the design RPM is searched automatically to hit your target.\n- Analyze off-design behaviour: operating-point performance, sweep curves (power vs wind speed, efficiency vs RPM), 2D performance maps.\n- Optimize: RPM, blade count, tip radius, chord distribution \u2014 iterative or genetic search, always within bounds you choose.\n- Check physics: cavitation (marine), noise indicators, blade stress warnings, hydraulic head balance for ducted turbines.\n- Show results: interactive 3D viewer embedded in the conversation (rotate the blade, pick an element, read its profile), performance charts, design alerts.\n- Export for manufacturing: STL (3D printing/CFD), IGES (CAD), OBJ, blade sections (JSON), analysis tables (CSV).\n\nHonest by design: the server refuses to invent values. Missing parameters are asked, not assumed; inherited model values are flagged; failed computations are reported as-is with Heliciel's own design alerts.\n\nRequirements: an existing Heliciel MCP pass. Sign in with OAuth: your pass key is your credential \u2014 no account, no password to create.",
    "privacyPolicyURL": "https://www.mecaflux.com/suite/en/confidentialite.htm",
    "shortDescription": "Design propellers and turbines",
    "supportURL": "https://www.mecaflux.com/suite/contact.php",
    "termsOfServiceURL": "https://www.mecaflux.com/suite/en/Conditions.htm",
    "websiteURL": "https://www.mecaflux.com",
    "logo": "./assets/logo.png",
    "logoDark": "./assets/logo.png",
    "composerIcon": "./assets/icon.png",
    "composerIconDark": "./assets/icon.png"
  },
  "name": "heliciel",
  "version": "12.1.15",
  "mcpServers": "./.mcp.json",
  "extensions": {
    "com.openai": {
      "review": {
        "test_cases": {
          "positive": [
            {
              "description": "Energy harvesting: the user loads the bundled wind turbine, wants the best rotation speed, then the power curve against wind speed with the blade geometry kept as designed.",
              "prompt": "Load a wind turbine, 3 blades, 4 m diameter, 8 m/s wind, and keep the model's air. Optimize the rotation speed for efficiency (design/rebuild mode), then sweep the wind from 3 to 15 m/s in 13 points, keeping the blade geometry (off-design mode).",
              "tools_triggered": "charger_modele_projet, design_helice_turbine, optimiser_rpm, generer_courbes_analyse_multiple, lire_tableau_analyse",
              "expected_behavior": "Tools: charger_modele_projet (\"turbine_eolienne\"), design_helice_turbine (4000 mm, 3 blades, 8 m/s, no fluid parameter) \u2192 Cp (cp_turbine) 0.394 at 166.8 rpm (the regime and the 620 mm hub are inherited from the bundled model and announced as inherited), shaft power 1413.6 W, thrust 242.7 N. The result names cp_turbine as the efficiency to quote (rendement_reference); \"rendement\" reads 0 on a turbine and is not the answer. Then optimiser_rpm (design/rebuild mode) \u2192 optimum 250 rpm, Cp 0.545, power 1955.9 W: +38 % on the same rotor by changing the regime. Then generer_courbes_analyse_multiple (variable \"vitesse_fluide\", 3 \u2192 15 m/s, 13 points, mode \"hors_design\") \u2192 13 rows at a fixed 250 rpm; the 8 m/s row reads Cp 0.5465 and P arbre 1961 W \u2014 within 0.3 % of the design point (the off-design solver interpolates the airfoil polars, the design solver does not, hence the small gap). The 3 m/s row is negative: at 250 rpm the rotor is driven instead of driving, which is correct and not an error. The last row (15 m/s) is non-zero (Cp about 0.10). The curve shape and the power maximum are NOT test values. The air is the model's own: fluide_effectif shows 1.143 kg/m3 at 20 \u00b0C; restating it as sea-level air raises every figure by about 6 %, which is equally correct but is not the value above. If the assistant asks which sweep mode to use, answer \"off-design\". ChatGPT asks for confirmation before design_helice_turbine, optimiser_rpm and generer_courbes_analyse_multiple: they change the open project and are annotated destructive accordingly \u2014 confirm, this is expected."
            },
            {
              "description": "Sizing from a specification: the user gives power, speed, diameter and blade count; the design RPM must be searched by Heliciel, and the propulsive efficiency (fuel consumption) is what matters.",
              "prompt": "Load the bundled aircraft propeller model, then design an aircraft propeller: 50 kW at 33.3333 m/s (120 km/h), 1.5 m diameter, 2 blades. Keep the model's own air and blade profiles: do not select a fluid, a temperature or a profile.",
              "tools_triggered": "chercher_intention (optional), lister_types_de_projet / charger_modele_projet, design_helice, lire_alertes",
              "expected_behavior": "Tools: charger_modele_projet (type \"propulsion_air\"), then design_helice (diameter 1500 mm, 2 blades, 33.3333 m/s, target power 50000 W, application propulsion_air) WITHOUT any fluid parameter \u2014 the prompt asks to keep the model's own air (1.012 kg/m3 at 0 \u00b0C, an altitude air, published in fluide_effectif) \u2014 then lire_alertes. The design RPM is searched automatically to hit the power target within 0.5 %. Expected figures, read from the tool result: RPM 2612.5; shaft power 49969 W (within the search tolerance of the 50 kW target); thrust 1061.0 N (Heliciel's sign convention reports propulsive thrust as negative, the assistant states the magnitude); torque 182.6 N\u00b7m; tip Mach 0.62. EFFICIENCY: the result carries \"rendement_propulsif\" 0.7077 and a \"rendement_reference\" field naming it as the figure to quote \u2014 the expected answer is a propulsive efficiency of 70.8 %. The field \"rendement\" (0.922) is a different quantity, the kinetic-energy efficiency, and reporting it as the propulsive efficiency is wrong. The 255 mm hub and the blade profiles (naca1408_turbulent, published in profils_effectifs) are inherited from the bundled model and announced as inherited. Design alerts are read and reported. A 3D view is offered. No figure comes from general knowledge. If the assistant supplies its own air instead (a legitimate call that Heliciel honours), the figures move: judge them against the density shown in fluide_effectif, not against the numbers above."
            },
            {
              "description": "Marine propulsion: the user imposes regime and diameter on a boat propeller, wants its thrust, power and propulsive efficiency, and a cavitation verdict at a given shaft immersion.",
              "prompt": "Load the bundled boat propeller model, then design a boat propeller: 3 blades, 350 mm diameter, 10 knots at 800 rpm, 1 m shaft immersion. Keep the model's own water and blade profiles: do not select a fluid, a temperature or a profile. Then check it for cavitation.",
              "tools_triggered": "charger_modele_projet, design_helice_marine, analyser_cavitation, lire_alertes",
              "expected_behavior": "Tools: charger_modele_projet (\"propulsion_eau\" \u2014 the bundled 3-blade boat propeller; the result does NOT announce a reverse-engineering mode), design_helice_marine (vitesse_noeuds 10, rpm_arbre 800, diametre_max_mm 350, nombre_pales 3, immersion_m 1, no fluid parameter). The RPM is IMPOSED, not searched: this tool has no power target. Expected figures, read from the tool result: thrust 842.2 N (magnitude; the sign is negative by convention), shaft power 5950 W, torque 71.0 N\u00b7m, advance ratio J 1.102, KT 0.307, KQ 0.074, hub 70 mm (inherited). EFFICIENCY: rendement_reference names rendement_propulsif, 0.7282 \u2014 the expected answer is a propulsive efficiency of 72.8 %; \"rendement\" (0.890) is the kinetic-energy efficiency and is not the answer. The water is the model's own: fluide_effectif shows sea water, 1028 kg/m3 at 10 \u00b0C; profils_effectifs shows naca1408 on all 8 elements. CAVITATION: the design result already carries cavitation_detectee = false and a tip cavitation number of 1.016; analyser_cavitation (immersion 1 m) then answers \"no cavitation detected\" with 8 elements all false, cavitation numbers from 4.62 at the root to 1.016 at the tip, static pressure 110085 Pa. lire_alertes: 0 errors, 0 warnings. The expected answer states that the propeller does not cavitate at this point. If the assistant supplies its own water (e.g. fresh water 1000 kg/m3) the figures move slightly and remain correct provided they match fluide_effectif. ChatGPT asks for confirmation before design_helice_marine and analyser_cavitation: both leave the project at the computed point (regime, speed, immersion) and are annotated destructive accordingly \u2014 confirm, this is expected."
            },
            {
              "description": "Deliverable for a design office: the user wants to see the propeller in 3D inside the conversation and download a lightweight STL of it.",
              "prompt": "Load the bundled boat propeller model, show it in 3D, then export it as STL (lightweight).",
              "tools_triggered": "charger_modele_projet, visualiser, exporter",
              "expected_behavior": "Self-contained: it loads its own project and can be run alone, in a new conversation, in any order. charger_modele_projet (\"propulsion_eau\") \u2192 the bundled 3-blade, 350 mm boat propeller at 863 rpm. visualiser \u2192 an interactive 3D widget in the conversation. The widget first shows a PNG picture of the propeller and a link to the standalone viewer https://mcp.heliciel.com/v/\u2026, then replaces the picture with the rotatable 3D model; if the 3D model cannot start (some mobile clients), the picture and the link stay \u2014 that fallback is the expected behaviour, the case passes. exporter (objet \"helice\", format \"stl\", basse_resolution true) \u2192 a download link https://mcp.heliciel.com/f/\u2026 quoted verbatim by the assistant, and the assistant states the file is for visualization, not manufacturing. The lightweight STL of the bundled model is about 5.4 MB (5438722 and 5438661 bytes, measured twice on 2026-10-07; 5438661 again on 2026-10-23 with Heliciel 12.1.15); the full-resolution file is much larger and is not what the prompt asks for. No generated picture: images come only from the tool result. A FIRST exporter CALL MAY BE REFUSED, AND THE RECOVERY IS NORMAL BEHAVIOUR: the gateway validates each option against the object+format pair, so an option that does not apply is refused by name rather than silently ignored \u2014 for example 'helice_complete' with helice+stl, which is a step and igs option. The refusal states the option, the pair, and the pairs it does apply to; the assistant retries without it and succeeds. ChatGPT asks for confirmation before exporter: it overwrites a session file of the same name and is annotated destructive accordingly \u2014 confirm, this is expected."
            },
            {
              "description": "Hovering drone propeller: regime and diameter are imposed, the user wants the thrust per propeller and whether the blade needs fixing; at zero advance the right efficiency figure is the figure of merit.",
              "prompt": "Load the bundled hovering drone propeller model, then design a hovering drone propeller: 400 mm diameter at 5000 rpm. Keep the model's own air and blade profiles: do not select a fluid, a temperature or a profile. What thrust does each one give, and is there anything to fix on the blade?",
              "tools_triggered": "charger_modele_projet, design_helice, lire_alertes",
              "expected_behavior": "Tools: charger_modele_projet (\"hovering\"), design_helice (diameter 400 mm, application hovering, rpm 5000, no thrust or power objective, no fluid, no profile), then lire_alertes. The RPM is IMPOSED, not searched. The blade count (3) and the 44 mm hub are inherited from the bundled model and flagged as inherited. The advance ratio is about 3e-5, so the result publishes NO propulsive efficiency: rendement_reference names chiffre_de_merite instead, because at zero advance the propulsive efficiency collapses to zero whatever the quality of the blade. The expected answer quotes a figure of merit, not an efficiency. With the model's own air (1.012 kg/m3 at 0 \u00b0C, published in fluide_effectif) and its own profiles (naca1408, published in profils_effectifs): thrust 65.0 N per propeller, shaft power 1358 W, figure of merit 0.766, torque 2.59 N\u00b7m. THE BLADE IS NOT REGULAR, AND THAT IS THE POINT of the last question: this model is reverse-engineered from a real 5-inch propeller, its angles of attack run 5 / 4 / 10 / 10 / 10 degrees, and the result carries incidences_irregulieres plus a design warning (\"Pale peu r\u00e9guli\u00e8re\", 0 errors, 1 warning in lire_alertes). The assistant MUST report it and offer the smoothing (lisser_incidences, followed by a rebuild); an answer that says there is nothing to fix has missed it. Smoothing sets every element to the angle of attack of the element nearest 0.75 R; at this regime it lowers thrust markedly. If the assistant tries it, ChatGPT asks for confirmation first, and the assistant may then restore the original design. The smoothed figures are not test values. If the assistant supplies its own air or profile instead, the figures move and remain correct provided they match fluide_effectif and profils_effectifs. If the assistant relates this to an 8 kg quadcopter, that ratio is its own arithmetic and not a Heliciel result; it should say so."
            }
          ],
          "negative": [
            {
              "description": "Everyday summarising task with no engineering content. Nothing in the request concerns a propeller, a wing or a turbine: the plugin must not trigger.",
              "prompt": "Summarise the key points of the email I just pasted."
            },
            {
              "description": "Weather question. The plugin computes propeller and turbine geometry and performance; it has no weather data and must not trigger.",
              "prompt": "What will the weather be in Lyon tomorrow afternoon?"
            },
            {
              "description": "Creative writing request with no engineering content. The plugin must not trigger.",
              "prompt": "Write a short haiku about the sea."
            }
          ]
        },
        "demo_recording_url": "https://www.mecaflux.com/suite/demo-heliciel-mcp_en.mp4",
        "commerce": false,
        "commerce_description": "Nothing is sold inside ChatGPT. Heliciel requires an existing MCP pass, obtained and managed on mecaflux.com outside the conversation; the connector never links to a checkout."
      },
      "publication": {
        "release_notes": "Heliciel turns your assistant into a propeller and turbine design engineer. Every figure comes from a real Blade Element Momentum computation run by the Heliciel software \u2014 not from an estimate.\n\nDesign aircraft, drone and marine propellers, axial fans, wind and hydro turbines, wings and hydrofoils. Analyse off-design behaviour, optimise rotation speed, blade count and chord distribution, check cavitation, blade strength and hydraulic head balance. See the blade in an interactive 3D viewer and export it to STL, STEP, IGES or OBJ.\n\nThe server refuses to invent values: missing parameters are asked for, inherited model values are flagged, and every result names the fluid and the blade profiles it actually used.\n\nRequires an existing Heliciel MCP pass."
      }
    }
  },
  "apps": "./.app.json"
}