mistral-ocr-markdown

Mistral OCR Markdown

Overview

Use the bundled TypeScript CLI to convert one local PDF/image file into one markdown file via Mistral OCR.

Workflow

1. Validate that the input file path exists and is a single local document.
2. Resolve API key from CLI flag or environment.
3. Run the CLI to upload the file, execute OCR, and write markdown output.
4. Use the generated markdown file in follow-up agent steps.

Run The CLI

Use npm from scripts/.

npm install
npm run build
node dist/cli.js <input-file> [--output <output-md>]

Examples:

node dist/cli.js ../sample_files/example.pdf
node dist/cli.js /any/path/to/example.pdf --output-dir ../out
node dist/cli.js ../sample_files/example.pdf --output ../out/example.md
node dist/cli.js ../sample_files/example.png --pages 0,1,2

API Key Configuration

Provide Mistral API key in this precedence order:

1. --api-key <key>
2. MISTRAL_API_KEY in environment
3. .env file (auto-loaded from --env-file or nearest .env in current/parent directories)

Prefer MISTRAL_API_KEY in environment for CI/automation and .env for local development.

Output Contract

Write one markdown file.

• Keep page order.
• Insert page separators between pages.
• Preserve OCR markdown as returned by the API.
• Default output path: ./ocr-output/<input-basename>.md (relative to current working directory).
• Use --output for an exact file path or --output-dir for a custom output folder.

Parameters

• --model <name>: OCR model (default mistral-ocr-latest)
• --output-dir <path>: output directory when --output is not specified
• --pages <csv>: zero-based page indexes
• --keep-uploaded-file: skip cleanup of uploaded file in Mistral storage
• --include-image-base64: request image payloads from OCR response

References

Read references/mistral-ocr-api.md when changing OCR request fields or response handling.