sarvam-mcp 0.2.3

Official Sarvam AI MCP server — STT, TTS, Translate, Transliterate, LLM, Vision OCR for Indic languages

sarvam-mcp

Official Sarvam MCP server. Exposes every public Sarvam API — STT, TTS, Translate, Transliterate, Language ID, Text Analytics, LLM (30B / 105B), Vision Document Intelligence, Pronunciation Dictionaries — as first-class MCP tools so any MCP-aware client (Claude Desktop, Claude Code, Cursor, Windsurf, Zed) can call Sarvam with zero boilerplate.

Cross-platform Python package: macOS, Windows, and Linux (Python 3.11+).

Quickstart

1. Get your API key

Sign up or log in at dashboard.sarvam.ai/key-management and copy your API key (sk_...).

2. Add to your MCP client

Paste this into your MCP config JSON:

{
  "mcpServers": {
    "sarvam": {
      "command": "uvx",
      "args": ["sarvam-mcp"],
      "env": {
        "SARVAM_API_KEY": "sk_..."
      }
    }
  }
}

Replace sk_... with your actual API key.

If you've installed via pip install sarvam-mcp, you can use the console script directly:

{
  "mcpServers": {
    "sarvam": {
      "command": "sarvam-mcp",
      "env": {
        "SARVAM_API_KEY": "sk_..."
      }
    }
  }
}

3. Config file locations

Client	Config path
Cursor	~/.cursor/mcp.json (macOS/Linux) · %USERPROFILE%\.cursor\mcp.json (Windows)
Claude Desktop	~/Library/Application Support/Claude/claude_desktop_config.json (macOS) · %APPDATA%\Claude\claude_desktop_config.json (Windows)
Claude Code	claude mcp add sarvam -- uvx sarvam-mcp (then set SARVAM_API_KEY env var)
Windsurf	Cascade settings → MCP servers
Zed	settings.json → context_servers

Alternative: credentials file

Instead of setting SARVAM_API_KEY in the JSON config, you can store it in ~/.sarvam/credentials:

api_key = sk_...

The server checks SARVAM_API_KEY env var first, then falls back to ~/.sarvam/credentials.

Install

# Option A: run directly (no install needed)
uvx sarvam-mcp

# Option B: install globally
pip install sarvam-mcp

Tools

All defaults below reflect the latest non-deprecated models.

Tool	What it does	Default model
sarvam_stt_transcribe	Audio file → transcript (5 modes)	saaras:v3
sarvam_tts_speak	Text → audio file	bulbul:v3
sarvam_tts_stream	Text → streamed audio	bulbul:v3
sarvam_translate	Cross-language text translate	mayura:v1
sarvam_transliterate	Script conversion	—
sarvam_identify_language	Language + script detect	—
sarvam_text_analytics	Typed Q&A over text	—
sarvam_llm_complete	Chat completions	sarvam-30b
sarvam_vision_extract	Document Intelligence	Sarvam Vision
sarvam_vision_job_status	Poll Document Intelligence job	—
sarvam_pronunciation_list	List pronunciation dictionaries	—
sarvam_pronunciation_get	Get a pronunciation dictionary	—
sarvam_pronunciation_create	Create a pronunciation dictionary	—
sarvam_pronunciation_delete	Delete a pronunciation dictionary	—

Configuration

Env var	Default	Description
SARVAM_API_KEY	—	Required. API key from dashboard.sarvam.ai.
SARVAM_API_BASE_URL	https://api.sarvam.ai	Override for testing/staging.
SARVAM_MCP_BASE_PATH	~/Desktop	Where audio/document files land.
SARVAM_AUDIO_OUTPUT_MODE	files	files | resources | both.

Two namespaces

The server exposes tools across two namespaces:

• sarvam_tools_* — runtime tools. Call Sarvam APIs to do things (transcribe, speak, translate, etc.).
• sarvam_code_* — builder tools. Help you write code that uses Sarvam: docs, endpoint shapes, language lists, code snippets, starter projects.

Development

uv venv && source .venv/bin/activate
uv pip install -e ".[dev]"
pytest -q
mcp dev src/sarvam_mcp/server.py