ChivoxMCP
Chivox MCP · Exam-grade speech AI · 100M+ users

Make realOne MCP call to ship exam-grade assessment

Scores, phoneme diagnostics, personalized practice — one call. 95%+ expert consistency, same engine as China's college-entrance oral exam.

Reply within 1 business day · Integration assessment
30s Live DemoNo signup required
chivox-mcp · live preview
End-to-end
语音输入 → AI 评测 → 温柔反馈AI 口语陪练教育智能体出海中文 AppK12 听说教学智能硬件伴读
30s full demo

User speaks → MCP calls Chivox → LLM crafts feedback. Click for the live demo.

9.2B/ yr evaluations·95%+expert match·185countries & regions · 100M+ users·1 lineMCP integration
Pick Your Path

Are you here to evaluate, or to integrate?

Managers see value. Devs see how to ship.

Path A · Manager view
I'm a product / project lead

30-min call to align scope; plan & quote within 1 week. Here are the 4 things worth putting in your proposal:

Procurement-ready

Must-have

Vendor credibility concern?→Exam-grade speech assessment engine · trusted by 1,000+ education projects · 95%+ expert match.

Score + Teaching ready

No ML team

Afraid of scores only, no loop?→Prompt templates included. Score → Diagnose → Correct → Drill, full loop.

1-week legacy upgrade

Zero rewrite

Worried about big rewrites?→One MCP config line. Cursor / Coze / Dify / Feishu / DingTalk compatible.

Compliance-ready for edu

Compliance

Edu / gov-grade data residency?→ISO 27001 certified. Full HTTPS/TLS encryption, audio destroyed immediately after assessment.

See public wins
Path B · Developer view
I'm a technical lead

MCP (recommended) or Function Calling. Real 6-step production pipeline:

1
Audio In
URL · base64 · PCM
2
ASR
Phoneme align
3
Multi-dim
acc / flu / rhy
4
Diagnosis
LLM reads JSON
5
Feedback
Correct + drill
6
Power AI
16 tools to LLM
IN→ENGINE→LLM LOOP
mcp.json
1-line setup
{ "chivox": { 
  "command": "npx",
  "args": ["-y", "@chivox/mcp"],
  "env": { "CHIVOX_KEY": "sk-..." }
}}
Try in 30sRead docs
Full specs
Live Demo

Understand Chivox MCP in 30s

No signup, no payment — record, and see MCP return scores + phoneme diagnosis + LLM-generated personalized practice.

Evaluation engine
Question types
Real API Response
Every field comes from production Chivox APIs, not fake mock data
Reusable LLM Prompts
Includes full system prompts and output samples, ready to copy into your app
See Full Flow in 30s
No signup required—see evaluate → diagnose → practice instantly
Try Demo Now View Docs
mcp.cloud.chivox.com/demo · en-word
LIVE
Reference Word · en.word.scorecoreType en.word.score
unique/juːˈniːk/
00:00.000:02 / 0:03PCM · 16k
82
Overall
78
Pron.
LLM Diagnosis · Secondary Analysis
⚠ /juː/ lip rounding too flat⚠ weak stress on 2nd syllable✓ /niː/ is accurate✓ final consonant /k/ is clear
LLM Generated · Word Correction
"Repeat 3 times: lift tongue for /j/, round lips for /uː/ → YOU-niːk, keep stress on the 2nd syllable..."
Chinese / English evaluation4 real task types coveredReusable LLM promptsFull flow in 30 seconds
How It Works

How one audio turns into smart feedback

Take an "AI IELTS speaking coach" as an example. Developers configure once, users get automatic scoring + feedback every time.

The LLM picks and calls tools via MCP — no routing or assembly glue on your side.

user-request.txtEnd User
// Frontend → LLM context · Two audio modes supported

user:
  Please listen to my self-intro,
  where do I have pronunciation issues?

// Mode A · Pre-recorded audio (mp3/wav · URL or base64)
audio:
  https://cdn.app.com/u123/intro.mp3

// ─── OR ───────────────────────────────────────────
// Mode B · Real-time streaming (browser MediaRecorder / mini-program / app)
stream:
  audio/pcm · 16k · 16bit · mono
  → 200ms chunks, streaming results back

refText:
  "Hello, my name is Lucy..."
Use Cases

Give AI real ears, or plug into products and backends

LLMs gain ears via MCP; class chats, custom apps / backends, and global products share the same oral-assessment stack — pick MCP or plain HTTP as needed.

92
AI Speaking Coach

Turn your LLM into a speaking coach

"Practice with me for 5 min" — agent auto-generates prompts, listens, scores, corrects. Conversation is evaluation.

  • Conversational practice, seamless eval
  • Adaptive difficulty
  • Pinpoint phoneme / stress / pause
Kids Reading Companion

Give storybooks, pens & robots real "ears"

Kid reads → device calls MCP → LLM gives fun feedback → parents get reports.

  • Phonics / word / sentence eval
  • Encouraging prompts and gentle correction for kids
  • Hybrid device + cloud deploy
MCPCursorClaudeCozeDifyDoubaoGPT
Developer Ecosystem

One config, any AI tool gains speech eval

Cursor / Claude / Coze / Dify — any MCP client, plug and play.

  • Dual protocol: stdio / HTTP
  • 16 tools out of the box
  • Native LLM tool calling
9.2B evals / yearExam-grade speech assessment engine185 countries & regions
Proof

Not slide-deck stats — real production projects

Every project below is publicly verifiable. Want yours here too? We reply within 1 business day.

Reply in 1 business dayIntegration assessmentTransparent tiered quote
Exam-grade speech assessment engine
Trusted by 1,000+ education projects
95%+ expert consistency
20 years iterated, 9.2B evals/year
100M+ users · 185 countries & regions
China exams + global apps in parallel
Recent updates
See full list at chivox.com
2026-03
Inner Mongolia · Sonid Left Banner
Chivox won the Inner Mongolia Sonid Left Banner senior high school experimental building vocational training room equipment procurement
2025-12
Jiangsu Industry Bureau
Listed as Jiangsu provincial "Specialized & Innovative" SME 2025
2025-11
Heilongjiang · Jixian
High-school English listening & speaking platform procurement
2025-11
Heilongjiang · Daqingshan
Compulsory-education smart infrastructure upgrade
2025-11
CVTE · Seewo
Strategic partner for smart-education solution
2025-10
Hubei · Chibi
Junior-high oral exam training system citywide rollout
ISO 27001 certifiedJiangsu "Specialized & Innovative" SMEPrivate deployment / VPC / data residency