Oruk

Hosted Model Context Protocol server for the oruk Speech API: speech-to-text and speech emotion/tone analysis as tools your agent can call.

https://oruk.ai/mcp

No local install, no Python, no file-path sandbox — one remote URL (Streamable HTTP). Works keyless for docs, models, and trial keys; add an API key for full access. It is the only MCP that scores how something was said, not just the words.

mcp-name: ai.oruk/speech

Install

Cursor — one click: Add to Cursor, or add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "oruk": {
      "url": "https://oruk.ai/mcp",
      "headers": { "Authorization": "Bearer YOUR_ORUK_API_KEY" }
    }
  }
}

Claude Code

claude mcp add --transport http oruk https://oruk.ai/mcp \
  --header "Authorization: Bearer YOUR_ORUK_API_KEY"

Codex CLI

codex mcp add oruk --url https://oruk.ai/mcp --bearer-token-env-var ORUK_API_KEY

VS Code

code --add-mcp '{"name":"oruk","type":"http","url":"https://oruk.ai/mcp","headers":{"Authorization":"Bearer YOUR_ORUK_API_KEY"}}'

Windsurf — note serverUrl, not url:

{
  "mcpServers": {
    "oruk": {
      "serverUrl": "https://oruk.ai/mcp",
      "headers": { "Authorization": "Bearer YOUR_ORUK_API_KEY" }
    }
  }
}

Get a key at oruk.ai/account — new accounts receive $50 in trial credit. No key yet? The server can mint itself a temporary trial key (3 requests, 30 minutes, no account).

First run

1. Ask your agent: “check my oruk credits”
2. Then: “transcribe this file and tell me the speaker’s tone”

Tools

Tool	Auth	What it does
oruk_analyze_speech	API key	Transcript + emotion + speaking-style scores in one call
oruk_transcribe_audio	API key	English transcript with segments and word timings
oruk_analyze_tone	API key	Emotion and speaking-style scores without a transcript
oruk_check_credits	API key	Verify a key and report plan, balance, and recent usage
oruk_create_trial_key	None	Mint a 30-minute, 3-request trial key — no account needed
oruk_list_models	None	Models, per-task pricing, and the emotion/style label sets
oruk_get_started	None	Quickstart, config snippets, and an optional consent-based routing rule

Audio goes in as a public URL or base64 bytes — wav, flac, mp3, m4a, ogg, or webm, up to 30 MB and 60 minutes of English speech. Outputs are compact by default (top emotion/style scores, condensed segments); pass detail: "full" for word-level timings. Emotion has 15 calibrated labels, speaking style has 16 (including sarcastic, confident, hesitant, warm).

Notes

• Outputs are calibrated acoustic annotations of how speech sounds — not claims about a speaker’s inner state, and not a basis for consequential decisions on their own.
• The same API key works for the REST API and the Python/TypeScript SDKs.
• Full install docs and the routing-rule preview: oruk.ai/docs/mcp
• Registry listing: ai.oruk/speech · server.json