mistral-ai-cli

name	mistral-ai-cli
description	Mistral AI CLI. OCR (PDFs local/URL), TTS (text→speech w/ voice cloning), STT (audio→text w/ diarization). List voices/languages. Config API key, base URL, model. Outputs markdown w/ YAML frontmatter.

Mistral AI CLI

Installation

npm install -g @tan-yong-sheng/mistral-ai-mcp

Commands

OCR: Extract text from PDF

mistral-ai ocr <pdf-path> [--model MODEL] [--table-format markdown|html]

Extract text from PDF → markdown w/ YAML frontmatter.

Input: Local file or HTTPS URL.

Output format:

---
model: mistral-ocr-latest
pages: 35
---
# Extracted markdown content

Full text with headers, lists, formatting preserved...

TTS: Text to speech

mistral-ai tts <text> [--voice-id ID | --ref-audio FILE] [--format FORMAT]

Voice cloning: --ref-audio (5-25s audio file) → custom voice. Formats: mp3, wav, pcm, flac, opus.

Output: Binary audio file speech_${timestamp}.${format} in current dir.

List TTS voices

mistral-ai tts voices [--json]

Fetches available voices from API. --json for structured output.

STT: Speech to text

mistral-ai stt <audio> [--realtime] [--diarize] [--language LANG]

--realtime → <200ms latency model. --diarize → speaker labels. --language → lang code (en, fr, es, etc).

List STT languages

mistral-ai stt languages [--json]

Supported: en, fr, es, de, it, pt, nl, pl, ru, zh, ja, ko, ar, hi, tr.

Examples

# OCR local PDF
mistral-ai ocr ./document.pdf > /tmp/output.md

# OCR from URL
mistral-ai ocr https://arxiv.org/pdf/2501.00001.pdf > /tmp/paper.md

# TTS w/ preset voice
mistral-ai tts "Hello world" --voice-id en_paul_neutral > /tmp/speech.mp3

# TTS w/ voice cloning
mistral-ai tts "Custom message" --ref-audio my_voice.wav > /tmp/cloned.mp3

# STT w/ diarization
mistral-ai stt meeting.wav --diarize --language en

# STT realtime (low latency)
mistral-ai stt stream.wav --realtime

# List voices as JSON
mistral-ai tts voices --json | jq '.[] | .voice_id'

# Batch OCR
for pdf in *.pdf; do
  mistral-ai ocr "$pdf" > "${pdf%.pdf}.md"
done

Output Formats

OCR: Markdown + YAML frontmatter (model, pages). Full text w/ formatting preserved.

TTS: Binary audio file saved to disk (mp3/wav/pcm/flac/opus).

STT: Transcribed text + detected language.

Use Cases

OCR: Extract PDFs for indexing, convert scanned docs, batch invoices/contracts, feed to LLMs, process papers.

TTS: Generate voiceovers, voice cloning, accessibility, audio content.

STT: Transcribe meetings, podcasts, voice commands, multilingual support.

---

Configuration

Set API Key

mistral-ai config api_key YOUR_API_KEY

Stores in ~/.mistral-ai/config.json. Or env var: export MISTRAL_API_KEY=YOUR_API_KEY.

Set Base URL

mistral-ai config base_url https://api.mistral.ai/v1

Defaults to https://api.mistral.ai/v1.

Set Default Model

mistral-ai config model mistral-ocr-latest

Show Config

mistral-ai config show

Env Vars

• MISTRAL_API_KEY - API key (required)
• MISTRAL_AI_CONFIG_DIR - Config dir (default: ~/.mistral-ai)
• MISTRAL_BASE_URL - API endpoint (default: https://api.mistral.ai/v1)
• MISTRAL_MODEL - OCR model (default: mistral-ocr-latest)

Priority (high → low):

1. Env var
2. CLI flag
3. Config file
4. Default

---

MCP Tools

ocr_pdf - Extract text from PDFs (local/URL)

• Input: sources (array of paths/URLs)
• Output: Markdown (one section per source)
• Mechanism: Local → upload → signed URL → OCR

tts_speech - Generate speech from text

• Input: text, voice_id or ref_audio_url, format
• Output: Audio file path (/tmp/speech_${timestamp}.${format})
• Voices injected into description (dynamic from API)

stt_transcribe - Transcribe audio to text

• Input: audio_source, realtime, diarize, language
• Output: Transcribed text + language
• Languages injected into description (15 supported)

---

Troubleshooting

"MISTRAL_API_KEY required"

• mistral-ai config api_key <key> or export MISTRAL_API_KEY=<key>

Custom endpoint

• mistral-ai config base_url https://custom.api.com/v1

Custom model

• Flag: mistral-ai ocr file.pdf --model mistral-ocr-latest
• Config: mistral-ai config model mistral-ocr-latest
• Env: MISTRAL_MODEL=mistral-ocr-latest mistral-ai ocr file.pdf

Help

mistral-ai