Structural analysis in Python

Two ways to run it

Eurocode 3 checks from a script

A plain model gives you forces. check_beam goes one step further: it adds a ULS combination and an EN 1993-1-1 member check over the span, so one call returns bending, shear, combined N+M and lateral-torsional buckling utilizations with the per-clause trace behind each.

from fers_core import check_beam

# 7.5 m simply supported IPE400 in S275, 12 kN/m characteristic, ULS factor 1.35
beam = check_beam(7.5, "IPE400", material="S275", udl=12_000, uls_factor=1.35)
beam.run_analysis()

for row in beam.unity_check_results():
    print(row["governing"]["demand"])
# 0.8002749049232664

Buckling lengths default to the full span for a single-member model, so if the compression flange is restrained along its length — by decking, by a slab, by purlins — that is a modelling decision you make explicitly rather than one the default quietly makes for you. The worked EC3 example shows the same check with every clause written out.

The thing a web form cannot do

Five sections, five solves, one loop:

import os
import requests

KEY = os.environ["FERS_API_KEY"]

for section in ["IPE300", "IPE330", "IPE360", "IPE400", "IPE450"]:
    response = requests.post(
        "https://ferscloud.com/api/sdk/check-beam",
        headers={"X-API-Key": KEY},
        json={
            "span_m": 7.5,
            "section": section,
            "material": "steel_S275",
            "support": "simply_supported",
            "udl": 12.0,                       # kN/m, characteristic
            "idempotency_key": f"sweep-{section}-7m5",
        },
        timeout=60,
    )
    response.raise_for_status()
    check = response.json()["check"]
    print(f"{section:<8} UC {check['governing_utilization']:.3f}"
          f"  governed by {check['governing_check']}")

That is the argument for scripting a structural calculation. The marginal cost of the sixth variant is one more entry in a list, not another twenty clicks — and the same loop works over spans, load cases, steel grades, or a CSV of members exported from your model. What it reports is which limit state governs and by how much; which section you then choose is your call, not the script's.

Pass a stable idempotency_key and reuse it on retries: a repeat with the same key replays the stored result instead of charging a second solve.

Call it from any language

curl -s https://ferscloud.com/api/sdk/solve \
  -H "X-API-Key: keyId.secret" \
  -H "Content-Type: application/json" \
  -d '{ "model": { /* FERS model JSON */ }, "idempotency_key": "run-123" }'

Endpoints: solve, validate, check-beam, create-beam, sections, schema, models and me, all under /api/sdk/. The full machine-readable spec is at ferscloud.com/api/openapi.json.

Two habits save a lot of grief. Call validate before solve — it is free and it catches broken references and misspelled keys that the solver would otherwise silently resolve to a default. And ask schema for the contract rather than guessing field names, for the same reason.

Use it from an AI agent

The same solver is exposed as a remote MCP server, so Claude, ChatGPT, Cursor and VS Code can call it directly:

{
  "mcpServers": {
    "fers": {
      "url": "https://ferscloud.com/api/mcp",
      "headers": { "X-API-Key": "keyId.secret" }
    }
  }
}

The clients, the one-click installers, the OAuth flow and the full tool list are on the MCP server page.

Prefer JavaScript? The same engine is published on npm as a WebAssembly build that solves in the browser with no server round-trip — see structural analysis in JavaScript.

Units, axes and sign conventions

Metres, newtons and newtons per metre in the Python layer; elastic modulus in pascals; deflections come back in metres, so multiply by 1000 for millimetres. The REST beam helpers take kN and kN/m, which is why the sweep above passes 12.0 where the local call takes 12_000.

The axis convention catches everyone once: i_z is the strong (major) axis and i_y the weak one. The full set of axis, sign and unit conventions is written out on the modeling conventions page, and it is the first thing to read if a result comes back an order of magnitude off.

What it costs

The Python package is free to install and the solver runs locally, with a 100-member ceiling on the free tier. Hosted solves — REST and MCP — are free for the first 100 successful solves per rolling week, then charged from a prepaid balance, or unlimited on Pro at €19.95 a month. Only successful solves are charged. See pricing.

Frequently asked questions