The deepctl mcp command runs an MCP (Model Context Protocol) proxy that
connects to Deepgram's developer API and exposes tools to AI code editors.
# Login to Deepgram (stores credentials in system keyring)
deepctl login
# Or set the environment variable
export DEEPGRAM_API_KEY="your-api-key"Add to your project's .mcp.json:
{
"mcpServers": {
"deepgram": {
"type": "stdio",
"command": "uvx",
"args": ["deepctl", "mcp"]
}
}
}Or with an explicit API key:
{
"mcpServers": {
"deepgram": {
"type": "stdio",
"command": "uvx",
"args": ["deepctl", "--api-key", "YOUR_KEY", "mcp"]
}
}
}Add to .cursor/mcp.json:
{
"mcpServers": {
"deepgram": {
"command": "uvx",
"args": ["deepctl", "mcp"]
}
}
}Add to ~/.codeium/windsurf/mcp_config.json:
{
"mcpServers": {
"deepgram": {
"command": "uvx",
"args": ["deepctl", "mcp"]
}
}
}Replace uvx with pipx run in any of the above configurations:
{
"command": "pipx",
"args": ["run", "deepctl", "mcp"]
}If running from a local clone of the repository:
{
"command": "uv",
"args": ["run", "dg", "mcp"]
}By default the MCP server uses stdio (for editor integration). You can also run it as an HTTP server:
# SSE transport
deepctl mcp --transport sse --port 8000