Mistral OCR in your flow: try it and integrate it 3 ways

What you'll get

You turn any document — photo, scan, handwritten note, table, even formulas — into clean Markdown (and LaTeX for equations) in a few seconds, right inside your workflow. By the end of this guide you'll know:

Clean text and tables in Markdown, ready to paste wherever you need them.
The certainty of knowing where the model is unsure (confidence score) and where to double-check.
The right way to integrate it: the chat, an automation or a few lines of code.
A starting point without installing anything: 9 free tries on our tool.

Try it now, before you choose

Before deciding how to integrate it, test it in 30 seconds on our tool: no account, no payment. In about 0.4 seconds you get clean markdown. You get 9 free tries.

Try it now on the AI Pratika tool

The AI Pratika OCR tool: the handwritten document on the left, the extracted text on the right in 0.4 seconds — Our tool in action: handwritten document on the left, text extracted in 0.4 seconds on the right.

In 3 micro-steps:

1 Upload the image (photo, scan, screenshot — even crooked is fine).
2 Click "Extract text".
3 Click "Copy markdown" and paste it wherever you need it.

For people who don't code

1

Mistral's chat

When our tool's 9 tries aren't enough, Mistral's chat is home for people who don't code: you upload a document and ask for the extraction, in plain English. Three steps.

1 On chat.mistral.ai open the "+" menu and choose "Upload Files": upload a photo, scan or PDF.

The «Upload Files» menu open in Mistral's chat, ready to upload the document — Open «+ → Upload Files» and choose the file.

2 With the document attached, write the prompt below and send it.

Document attached in Mistral's chat with the extraction prompt written below — Document attached + the prompt, ready to send.

PROMPT TO PASTE

Extract all the text, keeping tables or special symbols if present

3 Mistral gives you back the text in Markdown — formulas in LaTeX included. Copy it. Done.

The text extracted by Mistral in Markdown, with math formulas rendered in LaTeX — The result: text extracted in Markdown, formulas included.

For those who want to automate

2

Automation with n8n

Do you need to process lots of similar files? A ready-made n8n template connects Mistral OCR to your flow. It's not one-click: you import it, add your key and adapt it.

The n8n template «Document Parsing with Mistral OCR»: HTTP Request nodes connected to Mistral OCR in three examples — The n8n template «Document Parsing with Mistral OCR»: three ways to connect Mistral OCR to your flow.

1 Open the n8n template and click "Use for free" to import it.

2 Connect your Mistral API key in the node credentials.

3 Point the flow at your files (the template uses Google Drive as the source).

4 Run it: every document automatically becomes structured Markdown.

It's a community template: always check what it does before running it. Just like on our tool, the file gets sent to Mistral for processing.

For developers

3

In your code (JS/TS)

It's the same approach that runs in our tool: a single fetch call, zero dependencies. It runs server-side — the key must never reach the browser. The image is passed inline as a base64 data URI.

SERVER · JS/TS · ZERO DEPENDENCIES

// SERVER-SIDE — la chiave Mistral non arriva mai al browser
export async function estraiTesto(imageDataUrl) {
  const res = await fetch("https://api.mistral.ai/v1/ocr", {
    method: "POST",
    headers: {
      "Content-Type": "application/json",
      Authorization: `Bearer ${process.env.MISTRAL_API_KEY}`,
    },
    body: JSON.stringify({
      model: "mistral-ocr-latest",
      document: { type: "image_url", image_url: imageDataUrl },
      include_image_base64: false,
      image_limit: 1,
    }),
  });
  if (!res.ok) throw new Error(`Mistral OCR ${res.status}`);
  const data = await res.json();
  return data.pages?.[0]?.markdown ?? "";
}

imageDataUrl is a data:image/…;base64,… string. The returned Markdown can contain HTML: sanitize it before displaying it. API reference: docs.mistral.ai.

Why it's not your usual OCR

Classic OCR "sees" characters. Mistral OCR 4 understands the document: it classifies the blocks — headings, table cells, equations, signatures, charts — and for each region returns coordinates, block type and a confidence score that tells you where the model is unsure.

The numbers back it up: independent annotators evaluated over 600 documents in more than 12 languages, preferring OCR 4 with a 72% win rate; it scores 85.20 on OlmOCRBench and covers 170 languages, including Italian.

Chart of Mistral OCR 4's benchmarks, taken from the official presentation document — Chart taken from Mistral OCR 4's official presentation document.

Common mistakes to avoid

Trusting low-confidence figures without double-checking them: the OCR proposes, you validate.

Using it for medical diagnoses, legal rulings or high-stakes financial decisions.

In code, displaying the Markdown without sanitizing it: it can contain malicious HTML (XSS).

Forgetting that the file reaches Mistral for processing (on our tool it doesn't stay on our servers).

Mistral OCR isn't magic: it's an accelerator. You choose the path, it reads the document — the final judgment call stays yours.

Mistral OCR in your flow try it and integrate it 3 ways

What you'll get

Try it now, before you choose

Mistral's chat

Automation with n8n

In your code (JS/TS)

Why it's not your usual OCR

Common mistakes to avoid

Google Search Console inside Claude Code

Deploy From Scratch: From Code to Link

AI Glossary

Get Claude to Watch Your Videos

Git and GitHub for Vibe Coders

Token Counter