rftools.io Adds MCP: 197 Calculators for AI

What Is MCP and Why Should You Care?

The Model Context Protocol (MCP) is an open standard that lets AI assistants call external tools. Think of it as a USB port for AI — any tool that speaks MCP can plug into any AI assistant that supports it.

Claude Desktop, Claude Code, Cursor, and a growing list of AI-powered development tools all support MCP. When you add an MCP server, the AI gains new capabilities. In our case: 197 RF and electronics calculators, runnable with a single function call.

Here's why this matters: instead of the AI trying to remember formulas (and sometimes getting them wrong), it calls the same validated calculator code that runs on rftools.io. You know those times when you ask an LLM to calculate something and it confidently spits out a result that's off by 10 dB? Yeah, this fixes that. The AI doesn't approximate — it calls the actual calculator implementation.

The difference is like asking someone to recite the quadratic formula from memory versus handing them a calculator. One approach works reliably every time.

Getting Started

Install via npm

The fastest way to try it:

npx rftools-mcp

This starts the MCP server locally. AI tools connect to it over stdio. Nothing fancy, just a standard MCP server that exposes calculator functions.

Claude Desktop Setup

Add this to your claude_desktop_config.json (found in ~/Library/Application Support/Claude/ on macOS or %APPDATA%\Claude\ on Windows):

{
  "mcpServers": {
    "rftools": {
      "command": "npx",
      "args": ["-y", "rftools-mcp"]
    }
  }
}

Restart Claude Desktop and you're done. Claude now has access to every calculator on rftools.io. The first time you ask it to calculate something RF-related, you'll see it discover and use these tools automatically.

Claude Code Setup

claude mcp add rftools-mcp -- npx -y rftools-mcp

One command. That's it. If you're using Claude Code for development work, this gives you instant access to calculators right in your workflow.

Three Tools, 197 Calculators

The MCP server exposes three tools. Simple interface, powerful capabilities.

list_calculators

Browse all available calculators, optionally filtered by category. There are 13 categories: RF, PCB, Power, Signal, Antenna, General, Motor, Protocol, EMC, Thermal, Sensor, Unit Conversion, and Audio.

Example prompt: "List all antenna calculators"

This returns everything from dipole design to parabolic dish gain calculations. Useful when you're exploring what's available or can't remember the exact name of the calculator you need.

get_calculator_info

Get the full specification of any calculator — inputs with units and defaults, outputs, and the formula used. This is how the AI knows what parameters to provide.

Example prompt: "What inputs does the microstrip impedance calculator need?"

The response tells you it needs trace width, substrate height, dielectric constant, copper thickness, and frequency. Each parameter comes with its unit (mm, GHz, etc.) and typical ranges. Most engineers skip reading documentation until something breaks — but when the AI reads it for you, that's actually helpful.

run_calculation

Run a calculator with specific inputs and get results with units. Every response includes a link back to rftools.io where you can see the interactive version.

Example prompt: "Calculate microstrip impedance for a 0.3mm trace on 0.2mm Rogers RO4003C (Er=3.55) with 1oz copper"

You get back the characteristic impedance (probably around 85 Ω for those dimensions), effective dielectric constant, electrical length, and losses. All with proper units, all verified against the same code that powers the web calculators.

Real-World Example: Designing a Receiver Front-End

Here's a conversation that would be painful to do manually but takes seconds with the MCP server:

You: I'm designing a 2.4 GHz receiver. The antenna connects through a 50-ohm microstrip to an LNA. My PCB is 1.6mm FR4. What trace width do I need for 50 ohms, and what's my link budget for a 100m range with 0 dBm transmit power?

The AI calls run_calculation twice:

1. Microstrip impedance — iterates trace width to find the 50-ohm match (~2.9mm on 1.6mm FR4)
2. RF link budget — calculates free-space path loss, received power, and link margin at 100m

You get precise, unit-correct answers in seconds, with links to the interactive calculators for further exploration. The AI handles the unit conversions (it knows your 1.6mm substrate height matters for impedance), picks reasonable defaults for copper thickness, and presents results you can actually use.

Try doing this manually: you'd open a microstrip calculator, fiddle with trace width until you hit 50 Ω, write down that number, then open the Friis calculator, convert your range to kilometers or keep it in meters depending on which calculator you're using, make sure you've got the right antenna gains... it's tedious. With MCP, you just ask.

The link budget calculation at 2.4 GHz over 100m gives you about 80 dB of path loss. If your transmitter puts out 0 dBm and you've got 0 dBi antennas on both ends (dipoles in free space), you're receiving around -80 dBm. Add some realistic antenna gains and you might be at -65 dBm. Now you know what sensitivity your LNA needs to hit. That's the kind of back-of-the-envelope calculation that MCP makes trivial.

Why Not Just Ask the AI to Do the Math?

Large language models are impressive at reasoning but unreliable at arithmetic. They might:

• Use a simplified formula that omits copper thickness corrections
• Get unit conversions wrong (mils vs mm, dBm vs watts)
• Round at the wrong step and accumulate errors
• Confidently present a wrong answer

I've seen GPT-4 calculate microstrip impedance and forget that the substrate height matters. I've seen it mix up effective dielectric constant with relative permittivity. These aren't occasional mistakes — they're systematic problems with asking a language model to do math.

The MCP server calls the exact same calculator code that runs on rftools.io. These are validated implementations — Hammerstad-Jensen for microstrip impedance, Friis for link budgets, exact dB/linear conversions — not LLM approximations. When you calculate a noise figure cascade, it's using the proper Friis formula for noise, not some half-remembered textbook equation.

This matters more than you might think. A 1 dB error in a link budget calculation could mean the difference between a working system and one that drops packets at range. A 10% error in trace width could shift your impedance enough to cause reflections. You don't want "close enough" — you want the same answer you'd get from doing the calculation properly yourself.

What's in the Box

All 197 calculators across 13 categories:

That's a lot of calculators. The RF section alone covers most of what you'd need for a typical wireless design — impedance matching, S-parameters, noise analysis, link budgets. The PCB calculators handle trace sizing for current and impedance, differential pairs (important for high-speed digital), and via design including thermal vias.

Power electronics gets its own substantial section because switching converter design involves a lot of tedious calculations. Buck converter component selection, flyback transformer design, MOSFET loss analysis — all things you can do by hand but would rather not. The battery life calculator is particularly useful for IoT projects where you need to estimate how long a coin cell will last given your sleep/wake duty cycle.

Signal processing covers the usual suspects: filter design (Butterworth, Chebyshev, Bessel), ADC resolution and SNR calculations, FFT bin resolution. The PLL loop filter calculator is one I use regularly — designing a stable loop filter by hand is error-prone.

The antenna calculators are straightforward implementations of standard designs. You're not going to design a phased array with these, but for a simple dipole or patch antenna, they'll get you in the ballpark. The Yagi calculator is handy for directional antennas at VHF/UHF.

EMC calculations are often neglected until you fail a compliance test. Having quick access to shielding effectiveness, filter design, and ESD protection calculations means you can design with EMC in mind from the start instead of fixing it later with expensive rework.

For AI Tool Builders: llms.txt

We also publish machine-readable documentation at the well-known /llms.txt path:

• rftools.io/llms.txt — summary with API info and MCP setup instructions
• rftools.io/llms-full.txt — complete listing of all 197 calculators with inputs, outputs, units, and URLs

The llms.txt specification is an emerging standard for making websites machine-readable. If you're building an AI agent that needs RF or electronics calculations, these files give you everything you need.

The full listing is exactly what it sounds like: every calculator, every input parameter with its unit and valid range, every output with its unit, and the direct URL to the interactive version. It's formatted for machine parsing but human-readable if you want to browse it. Think of it as API documentation for AI agents.

This is useful if you're building custom AI tools or agents. You can point your LLM at the llms-full.txt file and it'll know exactly what calculators exist and how to use them. Or use the MCP server directly — same information, different interface.

Open Source

The MCP server is part of the rftools.io codebase. The calculator functions are pure TypeScript with no browser dependencies — they work identically whether called from a web browser, an MCP server, or a Node.js script.

This means you can audit the code if you want. You can see exactly what formula is being used for any calculation. You can verify that the microstrip calculator is using Hammerstad-Jensen, not some simplified Wheeler approximation. Transparency matters when you're relying on calculations for real designs.

Install it today:

npx rftools-mcp

Or add to Claude Desktop and start designing circuits with AI assistance. The setup takes two minutes. The time you save on the first complex calculation makes it worthwhile.