One command, you're set.
Finds every MCP app on your machine and connects it to Sarvam. ~30 seconds. macOS and Linux.
curl -fsSL https://mcp.sarvam.ai/install | bash
Transcribe audio, generate speech, translate text, understand documents, and chat with Sarvam-M — all from Claude Desktop, Cursor, or any MCP-compatible app.
Cursor
Claude Code
Claude Desktop
Windsurf
Zed
Two ways to install. Both auto-detect your MCP apps and connect them to Sarvam.
Finds every MCP app on your machine and connects it to Sarvam. ~30 seconds. macOS and Linux.
curl -fsSL https://mcp.sarvam.ai/install | bashPaste this prompt into Claude, Cursor, or ChatGPT and let your AI handle the setup.
Set up the Sarvam MCP server on this machine.
Repository: https://github.com/sarvamai/sarvam-mcp
Please do the following, in order:
1. CLONE
git clone https://github.com/sarvamai/sarvam-mcp.git ~/sarvam-mcp
cd ~/sarvam-mcp
(If the directory already exists, just `git pull` it.)
2. INSTALL
Prefer `uv` if available, else fall back to venv + pip:
if command -v uv >/dev/null; then
uv venv -p 3.11 .venv
uv pip install -e .
else
python3 -m venv .venv
.venv/bin/pip install -e .
fi
3. WIRE UP MCP CLIENTS
Detect and configure each installed client. Use the absolute binary path
~/sarvam-mcp/.venv/bin/sarvam-mcp. Don't clobber other servers — merge.
Client config paths:
• Cursor ~/.cursor/mcp.json
• Claude Desktop ~/Library/Application Support/Claude/claude_desktop_config.json (macOS)
~/.config/Claude/claude_desktop_config.json (Linux)
%APPDATA%\Claude\claude_desktop_config.json (Windows)
• Claude Code run: claude mcp add sarvam -- ~/sarvam-mcp/.venv/bin/sarvam-mcp
• Windsurf ~/.codeium/windsurf/mcp_config.json
• Zed ~/.config/zed/settings.json (under `context_servers`)
The new entry per JSON file:
{
"mcpServers": {
"sarvam": { "command": "<abs path>/.venv/bin/sarvam-mcp" }
}
}
4. (OPTIONAL) API KEY
If I gave you a Sarvam API key, write it:
mkdir -p ~/.sarvam && chmod 700 ~/.sarvam
printf "api_key = %s\n" "$KEY" > ~/.sarvam/credentials
chmod 600 ~/.sarvam/credentials
If not, skip — Cursor/Claude will prompt me for it on the first tool call.
5. REPORT
Tell me which clients were updated, where the repo lives, and whether the
key was written. Remind me to restart the MCP client. If anything fails,
show the exact error and the fix.The full Sarvam platform — speech, language, vision, and reasoning — ready to use from any MCP-compatible app.
Transcribe audio in Hindi, Tamil, Bengali, and 20+ more languages
Generate natural-sounding Indic speech — pick a voice, get audio
Translate between English and 22 Indic languages instantly
Chat with India's first hybrid reasoning model — multilingual, multimodal
Extract text, tables, and structure from scanned or handwritten documents
Identify the language and script of any text automatically