The Mistral Parse tool provides a powerful way to extract and process content from PDF documents using Mistral's OCR API. This tool leverages advanced optical character recognition to accurately extract text and structure from PDF files, making it easy to incorporate document data into your agent workflows.
With the Mistral Parse tool, you can:
- Extract text from PDFs: Accurately convert PDF content to text, markdown, or JSON formats
- Process PDFs from URLs: Directly extract content from PDFs hosted online by providing their URLs
- Maintain document structure: Preserve formatting, tables, and layout from the original PDFs
- Extract images: Optionally include embedded images from the PDFs
- Select specific pages: Process only the pages you need from multi-page documents
The Mistral Parse tool is particularly useful for scenarios where your agents need to work with PDF content, such as analyzing reports, extracting data from forms, or processing text from scanned documents. It simplifies the process of making PDF content available to your agents, allowing them to work with information stored in PDFs just as easily as with direct text input.
Usage Instructions
Integrate Mistral Parse into the workflow. Can extract text from uploaded PDF documents, or from a URL.
Actions
Mistral PDF Parser
Parse PDF documents using Mistral OCR API
Input
| Parameter | Type | Required | Description |
|---|---|---|---|
file | file | Yes | Normalized UserFile from file upload or file reference |
resultType | string | No | Type of parsed result (markdown, text, or json). Defaults to markdown. |
pages | array | No | Specific pages to process (array of page numbers, starting from 0) |
apiKey | string | Yes | Mistral API key (MISTRAL_API_KEY) |
Output
| Parameter | Type | Description |
|---|---|---|
pages | array | Array of page objects from Mistral OCR |
↳ index | number | Page index (zero-based) |
↳ markdown | string | Extracted markdown content |
↳ images | array | Images extracted from this page with bounding boxes |
↳ id | string | Image identifier (e.g., img-0.jpeg) |
↳ top_left_x | number | Top-left X coordinate in pixels |
↳ top_left_y | number | Top-left Y coordinate in pixels |
↳ bottom_right_x | number | Bottom-right X coordinate in pixels |
↳ bottom_right_y | number | Bottom-right Y coordinate in pixels |
↳ image_base64 | string | Base64-encoded image data; returned only when the hidden includeImageBase64 input is enabled |
↳ dimensions | object | Page dimensions |
↳ dpi | number | Dots per inch |
↳ height | number | Page height in pixels |
↳ width | number | Page width in pixels |
↳ tables | array | Separate table objects, referenced from the markdown via placeholders like [tbl-0.html]. Mistral populates these only when table_format is "markdown" or "html"; Studio never sets it, so tables stay inline in the markdown and this list is empty |
↳ hyperlinks | array | Array of URL strings detected in the page (e.g., ["https://...", "mailto:..."]) |
↳ header | string | Page header content. Mistral returns it only when extract_header is true (it defaults to false); Studio never sets it, so this is not returned |
↳ footer | string | Page footer content. Mistral returns it only when extract_footer is true (it defaults to false); Studio never sets it, so this is not returned |
model | string | Mistral OCR model identifier (e.g., mistral-ocr-latest) |
usage_info | object | Usage and processing statistics |
↳ pages_processed | number | Total number of pages processed |
↳ doc_size_bytes | number | Document file size in bytes |
document_annotation | string | Structured annotation data as JSON string (when applicable) |