MCP server for AI clients and agents
UnitConvertLab publishes a Model Context Protocol (MCP) server so language models can convert units against real data instead of guessing. Five read-only tools cover 466 units across 36 categories, plus the regional and historical catalogue with its provenance notes. Connect once and any MCP-capable assistant — Claude, ChatGPT, Cursor, Codex, or your own agent — can call it.
Endpoint
https://nawpscdfwillbgkbupeo.supabase.co/functions/v1/mcpTransport: MCP Streamable HTTP. Authentication: none — every tool reads only the public conversion catalogue. POST requests must send Accept: application/json, text/event-stream, which the official MCP SDKs do for you.
Tools
| Tool | Input | What it returns |
|---|---|---|
| list_categories | — | Enumerate every measurement category, standard and regional, with unit counts and base units. |
| list_units | category | List every unit in one category with its symbol, aliases and relationship to the category base unit. |
| search_units | query | Resolve a free-text unit name, symbol or alias (for example "klick", "tbsp", "°R") to a canonical unit. |
| convert_units | value, from, to | Convert a value between any two units in the same category, including affine scales and regional variants. |
| regional_unit_info | unit | Provenance for a historical or regional unit: era, region, variant spread, risk notes and cited sources. |
Connect a client
Claude, ChatGPT and other hosted assistants
Open connector or MCP settings, choose “Add custom server”, select the Streamable HTTP transport and paste the endpoint above. Leave authentication empty. The five tools appear as soon as the handshake succeeds.
Config-file clients (Cursor, Codex, local agents)
{
"mcpServers": {
"unitconvertlab": {
"type": "http",
"url": "https://nawpscdfwillbgkbupeo.supabase.co/functions/v1/mcp"
}
}
}Prefer plain HTTP? The same numbers are available as static JSON on the free data API, and as an embeddable widget from the embed builder.
Why an MCP server for unit conversion
Language models are unreliable arithmetic engines for measurement work. They hallucinate factors for anything outside the metric–imperial core, silently mix up US and imperial gallons, and reproduce round-number approximations for traditional units whose real values vary by region and by century. A conversion tool backed by a cited factor table removes that failure mode: the model resolves the unit, calls the tool, and reports a number it did not invent.
Every tool response is derived from the same definitions rendered on the website, so an assistant’s answer matches the page a human would land on. Regional and historical units additionally carry their era, region, variant spread and source citation, which lets an assistant say “the Egyptian feddan is 4,200.83 m²” and explain which authority defines it.
Read how each factor is sourced and verified on the methodology page, and see every published correction in the corrections log.
Limits and licence
- All tools are read-only. Nothing on this server writes, stores or logs your prompts.
- No key, no account and no published rate limit. Please be reasonable with automated fan-out.
- Currency rates are not part of the MCP catalogue — use the live currency tool on the site instead.
- Data is licensed CC BY 4.0 — credit UnitConvertLab with a link wherever results are displayed.
Frequently asked questions
What is the UnitConvertLab MCP server?
It is a Model Context Protocol endpoint that exposes the UnitConvertLab conversion catalogue as callable tools. Any MCP-compatible AI client — ChatGPT, Claude, Cursor, Codex — can connect and run conversions against the same data the website uses.
Do I need an API key or an account?
No. The server is public and read-only. There is no sign-up, no key and no quota. It only reads the published conversion catalogue, so there is nothing private behind it.
How do I add the MCP server to Claude or ChatGPT?
Open your client's connector or MCP server settings, choose "Add custom server" (Streamable HTTP transport) and paste the endpoint URL shown on this page. The client will list the five tools automatically once it connects.
Which transport does the server use?
MCP Streamable HTTP. POST requests must send Accept: application/json, text/event-stream, which every official MCP SDK does by default.
How accurate are the conversions returned over MCP?
The tools read the same factor tables that power the website and the JSON data API, so an MCP answer and a page answer can never diverge. Factors are sourced and cited on the methodology page, and every correction is logged publicly.
Can I use the results commercially?
Yes, under CC BY 4.0. Credit UnitConvertLab with a visible link wherever the data or conversion output is displayed.