One chart, end to end.
From paste to export.
Follow one fictional size chart through the three tools, the way you would in Claude. Every output on this page came from the live server. Long outputs are trimmed, and each cut is marked.
Connect once.
Bring one chart.
Add the server to your MCP client before you start. The setup page covers the Claude custom connector (tested on 15 Nov 2026), Cursor and MCP Inspector. The current pilot documentation does not offer ChatGPT setup.
https://mcp.sizeref.online/mcp
This walkthrough uses Claude with the connector switched on in the chat. Cursor works the same way. In MCP Inspector there is no assistant: you send the tool calls shown on this page yourself.
- Paste the chart and ask for an audit.audit_size_chart reports what is wrong or open. It changes nothing.
- Read the problems and questions.Each problem has a fix. Each question lists its allowed answers.
- Fix the table and answer.Use the brand's own facts. The tool guesses none of them.
- Normalize.normalise_size_chart returns SizeRef chart format v1.
- Export.export_agent_ready_chart writes JSON-LD, CSV, text and SizeRef JSON.
- Review and place it.Check the numbers, then put each format where it belongs in your workflow.
This server takes size charts, never a person's measurements, and keeps none of the charts you send. Use only chart data you are entitled to process.
It is for use in the US and India, and not in the EU, EEA or UK. Read the pilot terms and the privacy details before you connect.
Start with
a messy chart.
The chart below is fictional. It is made for this page, and it has the kind of gaps real charts often have.
Size,Chest (pit to pit),Length,Sleeve,Colour S,20,27,8,Navy M,21,28,8.5,Navy L,22,29,9,Navy XL,23,30,9.5,Navy 2XL,24 1/2,31,TBC,Navy
What is messy about it:
- No unit. Nothing says inches or centimeters.
- One cell says
TBCinstead of a number. - The chart says
2XL, but the product's size buttons sayXXL. - A
Colourcolumn that is not a measurement. - A fraction,
24 1/2. This one is fine: the tools read fractions.
What to say to your assistant
Paste the chart with the facts you already know: the caption, the product name and the size buttons.
Audit this size chart with SizeRef Size Charts. It is a fictional sample. Its caption reads "Fictional sample tee, garment measurements". The product is called Sample Tee, and its size buttons are S, M, L, XL and XXL. Size,Chest (pit to pit),Length,Sleeve,Colour S,20,27,8,Navy …the rest of the chart above…
Your assistant builds the tool call from your message, so its call may differ a little. This is the call we sent to capture the output in step 2.
{
"table": {
"csv": "Size,Chest (pit to pit),Length,Sleeve,Colour\nS,20,27,8,Navy\n…\n2XL,24 1/2,31,TBC,Navy",
"caption": "Fictional sample tee, garment measurements"
},
"variant_sizes": ["S", "M", "L", "XL", "XXL"],
"product_title": "Sample Tee"
}The csv string is trimmed here. The call sent the whole chart.
Problems, questions,
and what it read.
The audit returned this. It is shown in full.
{
"verdict": "needs_fix",
"results": [{
"index": 0,
"caption": "Fictional sample tee, garment measurements",
"verdict": "needs_fix",
"problems": [{
"severity": "error",
"code": "CELL",
"size": "2XL",
"area": "sleeve",
"text": "sleeve for size 2XL is not a number or a range.",
"fix": "Write the cell as a number such as 38, a range such as 38-40, or a fraction such as 38 1/2."
}],
"questions": [
{ "key": "unit", "allowed": ["in", "cm"] },
{ "key": "category", "allowed": ["Tops", "Bottoms", "Leggings", "Outerwear", "Dresses", "Other"] },
{ "key": "fit", "allowed": ["compression", "bodyfit", "fitted", "regular", "relaxed", "oversized"] },
{ "key": "stretch", "allowed": ["low", "medium", "high"] },
{ "key": "children", "allowed": [true, false] }
],
"ignored_columns": [{ "col": 4, "header": "Colour" }],
"label_mismatch": {
"chart_only": ["2XL"],
"variants_only": ["XXL"],
"suggested_aliases": { "2XL": "XXL" }
}
}],
"note": "SizeRef stored nothing from this request."
}How to read it
- verdict: needs_fix
- Something is wrong or still open. A pasted table is always at least needs_fix, because category and children are always asked.
- problems
- One error: the 2XL sleeve cell,
TBC, is not a number. The problem names the size and the column, and says how to fix it. - questions
- Five facts the table does not state. Each lists the answers the tool accepts.
- ignored_columns
Colouris not a measurement, so it is left out.- label_mismatch
- The chart's 2XL has no matching size button, and the XXL button has no matching row. The tool suggests the alias
2XL → XXL. It does not apply it for you. - note
- Each answer from a tool carries the note "SizeRef stored nothing from this request." A call refused before the tool runs, such as one whose fields do not match the tool's inputs, gets a short refusal without it.
What it read from the words
Two facts were not asked, because the chart states them. The caption says "garment measurements", so the kind is garment. The header says "pit to pit", so the chest was measured laid flat. Unit, kind and flat are read only from plain statements like these, never from the numbers.
The answers this chart needs
| Key | What it asks | Allowed answers |
|---|---|---|
unit | Inches or centimeters? | in, cm |
category | Product category | Tops, Bottoms, Leggings, Outerwear, Dresses, Other |
fit | The cut's fit | compression, bodyfit, fitted, regular, relaxed, oversized |
stretch | Fabric stretch | low, medium, high |
children | Is this a children's chart? | true, false |
Other charts can also be asked kind, flat or sizeBy. The full list is on the tools page.
Use the brand's facts.
Not guesses.
Fix the bad cell in the table itself, then answer each question. Take every answer from the brand's own words: the product page, the tech pack, or the person who made the chart. In this fictional case, the 2XL sleeve is 10 inches.
The server remembers nothing between calls. The table is sent again, with the answers, on every normalize call. If your assistant does not resend it, ask it to.
What to say to your assistant
The 2XL sleeve is 10, so replace TBC with 10. The numbers are in inches. Category Tops, not a children's chart, fit regular, stretch low. Map 2XL to our XXL button. Now normalize the corrected chart.
If you leave an answer out
You get the open questions back, not a chart. This is a normal answer, not an error. Here the call answered unit, category and children, but not fit or stretch.
{
"ok": false,
"questions": [
{ "key": "fit", "allowed": ["compression", "bodyfit", "fitted", "regular", "relaxed", "oversized"] },
{ "key": "stretch", "allowed": ["low", "medium", "high"] }
],
"note": "SizeRef stored nothing from this request."
}If an answer contradicts the table
It is refused, never applied. Here the call also sent "flat": false, but the header already says "pit to pit".
{
"ok": false,
"problems": [{
"severity": "error",
"code": "BAD_ANSWER",
"text": "The chart already states flat \"true\", so the answer \"false\" was not applied.",
"fix": "Answer only what the chart leaves open, using one of the allowed answers."
}],
"note": "SizeRef stored nothing from this request."
}The server marks this reply as an error. Drop the contradicting answer, or fix the table if the table is wrong, and call again.
One standard chart.
Nothing inferred.
With the cell fixed and every question answered, normalise_size_chart returns the chart in SizeRef chart format v1. This is the call we sent.
{
"table": {
"csv": "Size,Chest (pit to pit),Length,Sleeve,Colour\nS,20,27,8,Navy\n…\n2XL,24 1/2,31,10,Navy",
"caption": "Fictional sample tee, garment measurements"
},
"answers": {
"unit": "in", "category": "Tops", "children": false,
"fit": "regular", "stretch": "low",
"aliases": { "2XL": "XXL" }
},
"product_title": "Sample Tee"
}The csv string is trimmed here. The call sent the whole corrected chart.
{
"ok": true,
"chart_v1": {
"v": 1,
"sizes": ["S", "M", "L", "XL", "2XL"],
"rows": {
"S": { "chest": 20, "length": 27, "sleeve": 8 },
"M": { "chest": 21, "length": 28, "sleeve": 8.5 },
"L": { "chest": 22, "length": 29, "sleeve": 9 },
"XL": { "chest": 23, "length": 30, "sleeve": 9.5 },
"2XL": { "chest": 24.5, "length": 31, "sleeve": 10 }
},
"kind": "garment", "flat": true, "unit": "in",
"category": "Tops", "fit": "regular", "stretch": "low",
"children": false,
"aliases": { "2XL": "XXL" }
},
"sizing_chart_cm": {
"type": "garment",
"areas": ["chest"],
"sizes": ["S", "M", "L", "XL", "2XL"],
"rows": {
"S": { "chest": [101.6, 101.6] },
… M, L and XL trimmed, same shape …
"2XL": { "chest": [124.46, 124.46] }
}
},
"kept": null,
"note": "SizeRef stored nothing from this request."
}What changed
24 1/2is now24.5.- The
Colourcolumn is gone. Chest, length and sleeve stay. - The kind and flat came from the chart's own words. Unit, category, fit, stretch, children and the alias came from your answers. The sizes and numbers came from the table.
sizing_chart_cmholds the numbers used for sizing, in centimeters. The flat chest is doubled: 20 inches flat is 40 inches around, which is 101.6 cm. Length and sleeve are carried and exported, but only chest, waist and hips are used for sizing.- The alias is shown, never used for sizing.
Field by field, the format is on the chart format page.
JSON-LD, CSV,
and plain text.
Pass the chart from step 4 to export_agent_ready_chart. Add the product's name, brand and URL, so the JSON-LD is made for that one product page. This fictional product uses a reserved example address.
What to say to your assistant
Export that chart as JSON-LD, CSV and text. The product name is Sample Tee, the brand is Example Apparel, and the URL is https://example.com/products/sample-tee. Show me anything that was omitted.
{
"chart_v1": { …the chart_v1 from step 4… },
"formats": ["jsonld", "csv", "text"],
"product": {
"name": "Sample Tee",
"brand": "Example Apparel",
"url": "https://example.com/products/sample-tee"
}
}The reply holds omitted, note, jsonld, jsonld_note, csv and text. Each part is shown below, as it reads once the string is unpacked.
OMITTED, NOTES
"omitted": [], "note": "SizeRef stored nothing from this request.", "jsonld_note": "Optional: the theme can link each variant's size to the matching @id in this block. This step is not required."
Nothing was omitted: schema.org has a measurement type for chest, length and sleeve on a garment. Shoulders, for example, would be left out of the JSON-LD and listed here with the reason.
JSONLD
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@graph": [
{
"@type": "SizeSpecification",
"@id": "https://example.com/products/sample-tee#size-S",
"name": "Example Apparel Sample Tee - S",
"hasMeasurement": [
{
"@type": "QuantitativeValue",
"value": 40,
"unitCode": "INH",
"valueReference": "https://schema.org/WearableMeasurementChestOrBust",
"description": "chest measured flat, shown as full circumference"
},
{
"@type": "QuantitativeValue",
"value": 27,
"unitCode": "INH",
"valueReference": "https://schema.org/WearableMeasurementLength"
},
{
"@type": "QuantitativeValue",
"value": 8,
"unitCode": "INH",
"valueReference": "https://schema.org/WearableMeasurementSleeve"
}
],
"description": "Example Apparel Sample Tee, size S. chest measured flat, shown as full circumference."
},
… sizes M, L and XL trimmed, same shape …
{
"@type": "SizeSpecification",
"@id": "https://example.com/products/sample-tee#size-2XL",
"name": "Example Apparel Sample Tee - 2XL",
"alternateName": "XXL",
"hasMeasurement": [
{
"@type": "QuantitativeValue",
"value": 49,
"unitCode": "INH",
"valueReference": "https://schema.org/WearableMeasurementChestOrBust",
"description": "chest measured flat, shown as full circumference"
},
… length 31 and sleeve 10 trimmed, same shape …
],
"description": "Example Apparel Sample Tee, size 2XL. chest measured flat, shown as full circumference."
}
]
}
</script>Because of the trims, this excerpt is not valid JSON. Copy the block from your own export, not from this page.
- One
SizeSpecificationper size, each with an@idbuilt from the product URL. - The flat chest is written as the full measurement, 40 inches, with a note saying so.
INHis the unit code for inches. A centimeter chart usesCMT.- The alias appears as
alternateName. - There is no
Productnode, so it does not repeat what the theme already prints.
CSV
size,chest (in),length (in),sleeve (in) S,40,27,8 M,42,28,8.5 L,44,29,9 XL,46,30,9.5 2XL,49,31,10
TEXT
Sample Tee: size chart, garment measurements in inches S: chest 40 in (20 in flat), length 27 in, sleeve 8 in M: chest 42 in (21 in flat), length 28 in, sleeve 8.5 in L: chest 44 in (22 in flat), length 29 in, sleeve 9 in XL: chest 46 in (23 in flat), length 30 in, sleeve 9.5 in 2XL (product option XXL): chest 49 in (24.5 in flat), length 31 in, sleeve 10 in
The fourth format, sizeref_json, is the chart in format v1 with $schema at the top. Leave formats out to get all four.
Your workflow.
Your call.
The server does not publish anything, change your store or sync a catalog. You decide where each format goes. Look over the result before you use it: a chart it checks, tidies or exports can be wrong or incomplete.
| Format | What it contains | Where you can put it |
|---|---|---|
jsonld | A whole <script type="application/ld+json"> block for one product | The HTML of that product's page, once |
csv | One row per size, flat chest, waist and hips doubled to the full measurement | A spreadsheet or your own tools. A cell a spreadsheet could read as a formula is written with a leading ' first. |
text | One readable line per size, with the flat number shown too | A size guide, a product description or your own notes |
sizeref_json | The chart in SizeRef chart format v1 | Your own tools. It is the format the SizeRef Shopify plugin, now in testing, reads. |
What to say to your assistant
Compare the exported CSV with the chart I pasted at the start, size by size. List every number that changed and why.
Then check the answer yourself against the source chart. In this example, the numbers that change should be the doubled chest values and the 2XL sleeve you supplied.
One way on Shopify
Shopify is not required. Any store that can add a block to its product pages can use the JSON-LD. On Shopify, one way is a product metafield:
- Add a multi-line text product metafield, for example
custom.size_chart_jsonld. - Paste each product's own block into that product's metafield.
- Print it once in the product template, with a Custom liquid block that holds the line below.
{{ product.metafields.custom.size_chart_jsonld.value }}This is a suggestion, not a tested recipe. Check it on your own theme before you rely on it. SizeRef makes no promise about what any assistant or search engine does with the block.
Check it by hand.
Then ship it.
- Every number matches the source chart, apart from the flat doubling you expect.
- Every answer came from the brand's own facts, not a guess.
- The
omittedlist is read, and anything left out is fine to leave out. - Any warnings are read. An
IMPLAUSIBLEwarning means a value is outside the usual range: check the unit and the digits. - Size labels match the product's size buttons, or have an alias.
- One block per product page, made for that product. A block's
@idvalues use that product's URL, so it must not go on another product. - After publishing, view the page source. The block appears once and is not shown as text.
- Things the server cannot see are checked by hand: chart images, sizes in a picture that the product does not sell, and the page's existing structured data.
Stuck on a chart?
Send it over.
Email fitlens2@gmail.com with the chart (not a customer's data) and what you expected. Problem codes are explained on the tools page.