FERS MCP Server — structural calculations for AI agents

FERS puts a real finite-element solver behind the Model Context Protocol. Install the local server and Claude, Cursor, VS Code, Codex or ChatGPT desktop can solve beams, frames and trusses, run Eurocode EN 1993-1-1 steel checks and return a hand-calc trace — computed on your own machine. Free for non-commercial use up to 100 members; Pro, which covers commercial use, up to 10,000.

Local MCP server (stdio, Node.js 18 or newer)

npx -y @ferscloud/fers-mcp

Install the local server Claude or ChatGPT in a browser?

What is the FERS MCP server?

The Model Context Protocol is an open standard that lets AI assistants discover and call external tools. FERS Cloud exposes its FEM solver as an MCP server so any compatible AI client can run structural calculations on your behalf.

Once configured, you can simply ask your AI assistant:

Use FERS to solve a cantilever beam with a 10 kN point load at the free end, 6 m span, IPE 300 steel section.

The AI will automatically call the FERS solver tools, get results, and explain the displacements, reactions, and internal forces to you.

Unlike a plain language model, FERS actually calculates: every result comes from a finite-element engine that is validated against closed-form solutions and returns the working, not a guess.

How to calculate a beam or frame with AI

Connect your AI client to the FERS MCP server, then describe the structure in plain language: span, section, supports and loads. The agent builds a solver-ready model, runs the FEM solver on your machine, and reports deflections, reactions, internal forces and Eurocode utilization.

Check an IPE 400 in S275, 7.5 m simply supported span, 12 kN/m distributed load, to EN 1993-1-1 with FERS.

Use FERS to build a 2D portal frame — 8 m span, 4 m eaves height, HEA 240 columns, IPE 330 rafter — apply 5 kN/m snow load and report the governing utilization.

A member check returns the bending, shear, combined N+M and lateral-torsional buckling utilizations, clause by clause, and check_beam adds the beam's deflection under the characteristic load. A solve returns displacements, reaction forces and internal forces — every number traceable in the report.

See the full Eurocode 3 worked example Try it in the browser first — Ask FERS, no install needed

Quick start: connect in two minutes

No account and no key are needed on the Free tier, which is for non-commercial use.

Install Node.js 18 or newer — the server runs with npx, nothing else to install.
Add FERS to your AI client with one of the snippets below.
Optional, for Pro limits: generate an API key on your Profile page and set it as FERS_API_KEY.

Using Claude or ChatGPT in the browser or on a phone? Those apps cannot start a program on your computer — connect them to the hosted server instead.

Install the local server

One npm package, @ferscloud/fers-mcp, for every client that can start a program on your computer. The FERS engine runs inside it, so every solve happens on your machine and the results come straight back to your agent.

Add to Cursor Install in VS Code

The one-click installs add the server without a key, on the Free tier (non-commercial use). Add FERS_API_KEY to its env for Pro, which covers commercial use.

Claude Desktop

One click: download fers-mcp.mcpb from the latest release and double-click it. Or add this to claude_desktop_config.json (Settings → Developer → Edit Config):

{
  "mcpServers": {
    "fers": {
      "command": "npx",
      "args": ["-y", "@ferscloud/fers-mcp"],
      "env": { "FERS_API_KEY": "<your key, optional>" }
    }
  }
}

Claude Code

claude mcp add --transport stdio fers --env FERS_API_KEY=<your key> -- npx -y @ferscloud/fers-mcp

One command in your terminal. Leave out --env and the key for the Free tier.

Cursor (.cursor/mcp.json)

{
  "mcpServers": {
    "fers": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@ferscloud/fers-mcp"],
      "env": { "FERS_API_KEY": "<your key, optional>" }
    }
  }
}

VS Code / GitHub Copilot (.vscode/mcp.json)

VS Code uses a top-level servers key instead of mcpServers.

{
  "servers": {
    "fers": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@ferscloud/fers-mcp"],
      "env": { "FERS_API_KEY": "<your key, optional>" }
    }
  }
}

Codex (~/.codex/config.toml)

[mcp_servers.fers]
command = "npx"
args = ["-y", "@ferscloud/fers-mcp"]
env = { FERS_API_KEY = "<your key>" }

ChatGPT desktop

In ChatGPT desktop: Settings → MCP servers → Add server → STDIO, with the command:

npx -y @ferscloud/fers-mcp

Windsurf, Cline, Zed and any other MCP client that starts local servers use the same command.

Options

FERS_API_KEY optional. A FersCloud API key (Profile → API Keys). Without one every solve runs on the Free tier, for non-commercial use up to 100 members; a key from a Pro account covers commercial use and raises the limit to 10,000.
--out-dir <folder> optional. solve_model with detail "full" writes the complete result to a file in this folder and returns its path, instead of putting it in the conversation.

What leaves your machine

Your models never do. With FERS_API_KEY set, the server only asks FersCloud for a short-lived licence token; without a key nothing is sent at all.

Claude or ChatGPT in a browser: the hosted server

Claude on the web and in the mobile app, and ChatGPT on the web, cannot start a program on your computer, so they connect to the hosted server. It builds and validates models like the local server but does not run the solver: solve_model, run_unity_checks and check_beam return a link that solves the model in your browser and shows the results there. The link is public and valid for 24 hours. For results inside the chat, use the local server.

Hosted endpoint (streamable HTTP, OAuth or X-API-Key)

https://ferscloud.com/api/mcp

Claude (web & mobile)

In Claude on the web or the mobile app: Settings → Connectors → Add custom connector, then paste the endpoint URL. OAuth signs you in — no key needed.

ChatGPT (web)

ChatGPT on the web adds it as a custom MCP server, which becomes a plugin — see the ChatGPT section below.

Other clients over HTTP

Any client that speaks streamable HTTP can use the hosted server too, for example where installing a local program is not allowed. Send your API key as the X-API-Key header; solves come back as a browser link.

Claude Code

claude mcp add --transport http fers https://ferscloud.com/api/mcp --header "X-API-Key: YOUR_API_KEY"

VS Code / GitHub Copilot (.vscode/mcp.json)

{
  "servers": {
    "fers": {
      "type": "http",
      "url": "https://ferscloud.com/api/mcp",
      "headers": {
        "X-API-Key": "your-api-key-here"
      }
    }
  }
}

Cursor (.cursor/mcp.json)

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

Windsurf, Cline, Zed and any other streamable-HTTP MCP client use the same URL and X-API-Key header.

Claude.ai: results in the chat, with the FERS skill

Claude on the web and in the desktop and mobile apps can run Python in its own code sandbox. With the FERS skill uploaded, Claude installs FERS there (pip install FERS) and solves your model inside the chat: deflections, internal forces, buckling and EN 1993-1-1 checks. The model goes to Claude's sandbox, not to FERS.

Download fers-skill.zip.
In Claude: Settings → Capabilities, turn on code execution, then under Skills upload fers-skill.zip.
Ask, for example: “Check an IPE200 in S235, simply supported over 6 m under 10 kN/m, to EN 1993-1-1.” Claude installs FERS in its sandbox and solves it there.

Download fers-skill.zip

The skill is free. The solve follows the engine's licence: free for non-commercial use up to 100 members; commercial use needs a Pro plan.

The tools at a glance

A selection of the tools; your client lists them all. Tool names are stable identifiers, so your agent discovers them automatically. The solve tools run the FEM engine on the local server and return a browser link on the hosted one. Saving to your FersCloud account requires Pro; the local server saves to a folder on your machine instead.

solve_modelsolve

Run the FEM solver on a structural model JSON (displacements, forces, reactions)

validate_modelfree

Check model structure without running the solver

run_unity_checkssolve

Solve and return Eurocode code-check (unity) results and report

check_beamsolve

Turnkey EN 1993-1-1 (EC3) steel member check — build, solve, and return the utilization (bending, shear, N+M, LTB) and the characteristic deflection

create_beamfree

Generate a solver-ready beam model (span, section, supports, load)

create_framefree

Generate a solver-ready 2D portal-frame model

build_modelfree

Build a model of any shape from a node and member list

set_analysisfree

Switch on second-order, modal, buckling or seismic analysis

add_unity_checkfree

Author a code check against the solver's canonical schema

get_check_schemafree

Fetch the canonical schema for authoring code checks

get_schemafree

Look up the model schema the solver is generated from, instead of guessing field names

list_sectionsfree

Browse the named European, UK and US steel sections

render_reportlocal only

Render the HTML hand-calc report of an already-solved model

save_modelPro on hosted

Store a model — in your FersCloud account on the hosted server (Pro), in a folder on your machine with the local server

load_modelfree

Load a previously saved model

list_modelsfree

List your saved models

list_templatesfree

Browse starter models

get_templatefree

Fetch a starter model

get_usagefree

Check your tier, its member limit and where solves run

Using ChatGPT? Desktop and the web

ChatGPT desktop runs the local server (Settings → MCP servers → Add server → STDIO, see above). ChatGPT on the web adds the hosted server as a custom MCP server:

Open chatgpt.com/plugins, press +, then Add custom MCP server
Name it FERS, enter the URL https://ferscloud.com/api/mcp and choose OAuth as the authentication
Create it as a plugin, install it, and type @ in a new chat to pick FERS

OAuth signs you in with your FersCloud account, with no key to paste. On the web, solves come back as a link that solves the model in your browser. A Custom GPT can still import https://ferscloud.com/api/openapi.json as an Action instead.

How much does it cost?

Nothing per solve. Every solve runs on your own machine or in your browser, so there is no solve quota and nothing is metered. Your tier only sets how large a model can be.

Tier	Model size	Price
Free	Up to 100 members	Free for non-commercial use — the local server needs no account or key
Pro	Up to 10,000 members	€19.95 / month — set FERS_API_KEY to a key from your Pro account

Saving models to your FersCloud account is Pro-only. On the hosted server, a browser link solves at the tier of the account that made it, for that model only.

Can you trust AI structural calculations?

Yes — when the AI does not do the arithmetic. FERS results come from a finite-element engine validated against closed-form analytical solutions, and every solve returns a transparent per-clause report you can check by hand. The LLM interprets; the solver calculates.

See the accuracy benchmarks Modeling conventions & units

Frequently asked questions

What is the FERS MCP server?

A Model Context Protocol server that lets AI assistants run real finite-element structural calculations — beams, frames and trusses, including Eurocode EN 1993-1-1 steel checks — and return the results with a hand-calc trace. It comes as a local server, the npm package @ferscloud/fers-mcp, that solves on your machine, and as a hosted server at https://ferscloud.com/api/mcp for chat apps that cannot start a local program.

Which AI clients can use FERS?

Any client that can start a local MCP server: Claude Desktop, Claude Code, Cursor, VS Code with GitHub Copilot, Codex, ChatGPT desktop, Windsurf, Cline and custom agents. Claude and ChatGPT in the browser or on a phone connect to the hosted server, which hands back a link that solves the model in your browser.

Do I need an API key?

Not for non-commercial use of the local server: without a key it solves on the Free tier and sends nothing anywhere. Commercial use, and models over 100 members, need FERS_API_KEY set to a key from a Pro account (up to 10,000 members). The hosted server signs in with OAuth (the Claude and ChatGPT connectors) or an X-API-Key header.

How much does the FERS MCP server cost?

There is no per-solve price and no solve quota: the local server solves on your machine, and the hosted server's links solve in your browser. Free covers non-commercial use with models up to 100 members; Pro, at €19.95 a month, covers commercial use and models up to 10,000 members, and adds cloud model storage.

Does my model leave my computer?

Not with the local server. The model is solved inside the server process on your machine; with FERS_API_KEY set, the only request to FersCloud asks for a licence token. The hosted server does receive the model, to validate it and park it behind the 24-hour browser link.

What can FERS calculate?

3D beams, frames and trusses with linear and nonlinear analysis: deflections, reaction forces, internal force diagrams, buckling, and Eurocode EN 1993-1-1 member checks including lateral-torsional buckling.

How accurate are the results?

The solver is validated against closed-form analytical solutions and published benchmark problems — the benchmarks page shows the comparisons. Every result includes a report you can verify line by line.

Should my agent use the MCP server or the command line?

Both run the same engine on your own machine. Desktop and editor clients (Claude Desktop, Cursor, VS Code, ChatGPT desktop) use the local MCP server: one tool call returns the checked numbers. An agent with a shell, such as Claude Code, Codex or a CI job, can use either: the MCP server answers a turnkey check in fewer turns, and the fers command from pip install FERS suits scripted work, with results kept in files. Chat apps in a browser or on a phone use the hosted server, which builds the model and returns a link that solves it in your browser. The terminal guide at ferscloud.com/docs/command-line has the working loop and a skill file.

More ways to use FERS

REST API & OpenAPI spec Ask FERS — chat demo in the browser Getting started with FERS Cloud