How to Use Mistral document OCR in Codex
Mistral document OCR extracts text, tables, and embedded images from PDFs and images, with optional structured JSON annotations. Use it to turn reports and scanned documents into content an agent can search, compare, or analyze.
Prompts to try with Mistral document OCR in Codex
Copy an example into Codex, or ask in your own words.
Use Mistral document OCR on the Mistral 7B paper at https://arxiv.org/pdf/2310.06825 and save its extracted text and images in this project's research directory.
Use Mistral document OCR on Apple's 2024 annual report PDF and build a CSV from its consolidated statements of operations, keeping the year headings attached to each column.
Use Mistral document OCR on Attention Is All You Need at https://arxiv.org/pdf/1706.03762 and create a Markdown research note from the extracted methods and results text.
Use Mistral document OCR on https://upload.wikimedia.org/wikipedia/foundation/f/f6/WMF_Mid-Year-Financials_08-09-FINAL.pdf and save the extracted financial tables as HTML fixtures in this project.
Use Mistral document OCR on the first eight pages of Microsoft's 2024 annual report PDF and save structured JSON containing the document title, reporting period, and a short summary.
How to set up Mistral document OCR in Codex
Pick the way that suits you. Each takes about two minutes.
Use Spiral's MCP server
- Add the hosted server
codex mcp add spiral --url https://api.spiral.ai/mcp
- Sign in
Codex opens the browser: sign in to Spiral and click Authorize. No API key. codex mcp list then shows Spiral.
codex mcp login spiral
Use the Spiral skill
- Send Codex this message
Your agent reads Spiral's skill and sets itself up.
set up https://spiral.ai/SKILL.md
- Sign in or paste a key
Approve Spiral in the browser, or give the agent a key from the console.
Use the Spiral CLI
- Install and sign in
Installs the Spiral CLI, signs you in and adds its skill to your coding agents.
npx -y spiral-cli@latest init
Call the Spiral API
- Create an API key
In the console, under API keys.
- Run Mistral document OCR
One request; the result comes back by polling or a webhook.
curl https://api.spiral.ai/v1/runs \ -H "Authorization: Bearer $SPIRAL_API_KEY" \ -H "Content-Type: application/json" \ -H "Idempotency-Key: $(uuidgen)" \ -d '{ "provider": "mistral", "model": "mistral-ocr-3", "input": { "document": { "document_url": "<document_url>" } } }'
Then paste one of the prompts above: Codex finds Mistral document OCR on Spiral, shows the price and runs it.
Mistral document OCR providers and prices
Spiral runs Mistral document OCR through Mistral. Open a card for today's price, with every fee included.
Settings you can ask for
Mistral document OCR works from your document annotation prompt. Ask Codex for any of these in your prompt.
| Setting | Allowed values | Default | Description |
|---|---|---|---|
| Image limit | a whole number | Max images to extract | |
| Table format | markdown or html | ||
| Extract footer | on or off | off | Extract the page footer into the response's |
| Extract header | on or off | off | Extract the page header into the response's |
| Image min size | a whole number | Minimum height and width of image to extract | |
| Include blocks | on or off | on | Return paragraph-level bounding boxes for all content blocks in the response |
| Include image base64 | on or off | Include image URLs in response | |
| Confidence scores granularity | word, page, or block | Granularity for confidence scores: 'page' (aggregate only), 'word' (per-word scores), or 'block' (per-block scores). Defaults to None (no confidence scores) to keep response payload small. |
Every API field of Mistral OCR 3 is on its tool page: open its card above.
What can you do with Mistral document OCR in Codex?
Build a searchable report collection from PDF text, reconstructed tables, and extracted images. A coding agent can turn the results into project data files and add structured annotations for document summaries.
Make annual reports searchable
Extract report text into Markdown for a company research index, then use it to find business segments, risks, and accounting notes.
Recover financial tables
Extract tables from public company reports for spreadsheet analysis. Preserve merged headers so year labels stay associated with the right figures.
Prepare research papers for analysis
Turn paper PDFs into extracted text and images that an agent can use to summarize methods, compare results, and inspect figures.
Compare public document revisions
Extract two editions of a government report or business publication, then compare the text and tables to identify changes.
Build a report metadata catalog
Use structured annotations to collect document titles, reporting periods, and summaries in a consistent JSON structure. Process long reports in sections of up to eight pages.
Describe figures for document search
Annotate extracted charts and figures with descriptions that help users find the relevant visual inside a collection of public reports.
More on Mistral document OCR and Codex
Overview
Mistral document OCR in other agents
Questions about Mistral document OCR in Codex
Can Codex use Mistral document OCR?
Yes. Add Spiral to Codex as an MCP server, and Codex can find and run Mistral document OCR like any built-in tool.
How much does Mistral document OCR cost in Codex?
$0.0026 per page ($0.0039 with document or bounding-box annotations). You pay from your Spiral balance as you go, with no subscription, and a failed run is not charged. Your Codex plan is separate.
Do I need a Mistral account to use Mistral document OCR in Codex?
No. Spiral runs Mistral document OCR on its own Mistral account. You sign in to Spiral once and never paste a key.
How long does it take to set up Mistral document OCR in Codex?
About two minutes: add the hosted server, then sign in.
Does Codex wait while Mistral document OCR runs?
Mistral document OCR runs in the background. Codex starts the run, checks on it and gives you the json when it's ready.
Where does the extracted information come from, and how current is it?
Mistral extracts content from the PDF or image supplied for processing. Its information reflects that document, so processing an older annual report gives you that report's contents.
Can Mistral preserve complex tables?
OCR 3 supports reconstructed tables with merged cells, multi-row blocks, and hierarchical column headers. HTML output can represent cells spanning multiple rows or columns.
How much of a document can structured annotations cover?
Mistral's cookbook documents an eight-page limit for document annotations and recommends splitting longer documents. The annotation stage receives OCR text plus the first eight extracted image bounding boxes.
Does the reported 74% benchmark mean 74% transcription accuracy?
No. Mistral reports a 74% overall win rate against OCR 2 across forms, scans, complex tables, and handwriting; that measures comparative wins rather than the percentage of characters transcribed correctly.
Does OCR 3 return paragraph bounding boxes?
Paragraph-level block extraction requires OCR 4 or newer. Mistral says older models accept the option but return an empty array, so OCR 3 should not be used when paragraph coordinates are required.
Start using Mistral document OCR in Codex
One Spiral account for every tool your agent needs. No subscription; you pay per run.
Paste this into Codex:
set up https://spiral.ai/SKILL.md


