Integration comparison

Same assessment engine.Three ways to own the call.

Choose by who captures the audio, who decides when to score, and how much of the orchestration your team wants to own—not by the scoring result you need.

Same English & Mandarin modelsSame structured evidenceDifferent integration contract

Three devices on a desk showing SDK, MCP, and function-calling speech assessment UIs

/quick answer

Start with the owner of the experience.

Speech assessment SDK on a laptop and phone with waveform scoring UI

Choose this for an app

SDK & API

You own the product experience

Best when your app captures audio and needs scoring embedded directly in the learner journey.

Native learner-facing UI
Streaming and submitted audio
Enterprise licensing

Explore SDK & API

Bright desk with MCP Inspector tools and speech assessment results on a laptop

Choose this for an MCP agent

MCP server

The agent discovers the tools

Best when an MCP-compatible client should choose and call speech assessment tools itself.

Standard MCP tool discovery
16 English and Mandarin tools
Self-serve evaluation points

Explore MCP server

Choose this for a custom agent

Function calling

You own the orchestration

Best when your existing agent stack should decide when and how an assessment function runs.

Your schemas and tool loop
REST or WebSocket transport
Same pricing as MCP

Explore function calling

/decision matrix

Compare the choices that change your build.

Scoring depth is intentionally absent from this table: accuracy, fluency, integrity, word and phoneme evidence are shared engine capabilities.Swipe horizontally to compare all three paths.

Decision	SDK & API	MCP server	Function calling
Fit & ownership
Best for	Learner-facing apps and education platforms	MCP clients, IDE agents and assistants	Custom agents and workflow backends
Who controls the experience	Your product code and interface	The MCP client and model	Your orchestrator and model
Who decides to score	Your application logic	The model, from the discovered tool list	The model, from schemas you provide
Implementation
Runs in	iOS · Android · web · server	Any Streamable HTTP MCP client	Any backend or agent framework
Transport	Native SDK, REST and WebSocket	MCP over Streamable HTTP	REST or WebSocket
Authentication	Application ID + signed timestamp	Bearer API key	Bearer API key
Audio handling	Capture, stream or upload in your app	Pass audio through a tool argument	Pass audio through your function handler
Commercial model
Getting started	Talk to Chivox about product requirements	Create an account and API key	Create an account and API key
Pricing	Enterprise licensing via BD	Self-serve evaluation points	Self-serve evaluation points

/same request, three wrappers

The contract changes. The assessment does not.

Each path sends the same language, reference text and recording. What changes is where the call is declared and who executes it.

One shared speech assessment dashboard shown across monitor, laptop, and phone

Your app starts the session

SDK & API

engine.start({
  coreType: 'en.sent.score',
  refText: 'An apple a day.'
});

The model calls a discovered tool

MCP server

mcp_assess_pronunciation({
  language: 'en-US',
  ref_text: 'An apple a day.',
  audio_url
});

Your handler executes the schema

Function calling

assess_speech({
  language: 'en-US',
  ref_text: 'An apple a day.',
  audio_url
});

Shared structured response

The same evidence reaches the next product decision.

Overall dimensions, word alignment, phoneme detail and supported Mandarin tone evidence remain available regardless of the wrapper.

Accuracy 82Fluency 85Integrity 95

Speech assessment product moment on laptop and phone

Still deciding?

Start from the product moment, not the protocol.

Show us who records, who acts on the score and where the feedback appears. We can map that workflow to the right integration path.

Choose a live demo MCP & Function Calling

/contact

Let’s build a better speaking experience together.

Tell us what you’re building and whether you need SDK/API, MCP, or function calling. We’ll reply within one business day with pilot credits, pricing, or a deployment plan.