Frequently Asked Questions
Most frequently asked questions are collected here. Still missing an answer? Talk to us directly—from protocol details to pricing plans.
MCP stands for Model Context Protocol — a standardized protocol that exposes external tools and data sources to large language models in a unified format. Chivox MCP uses this protocol to provide structured speech assessment data for LLM consumption.
Users send audio through a client. The Chivox MCP server completes the speech assessment and returns structured data with multi-dimensional parameters like overall score, accuracy, and fluency. This data can be directly understood by LLMs to generate teaching feedback.
Currently we support Chinese (Mandarin) and English speech assessment. More languages are being added — please contact us for specific language needs.
Install the @chivox/mcp package via npm (`npx -y @chivox/mcp`), add one line to the mcp.json of Claude Desktop / Cursor with your API Key, and you're ready. See the developer docs for details.
Traditional APIs only return scores that developers must interpret and present. MCP outputs assessment data in an LLM-readable format, allowing models to directly perform secondary analysis and generate teaching content — significantly reducing development costs.
MCP services use HTTPS / TLS 1.2+ encrypted transmission. Audio is destroyed immediately after assessment with zero retention. Chivox is ISO 27001 certified and China MLPS Level 3 assessed.
Yes. MCP provides streaming assessment capabilities, returning intermediate results while the user is reading, suitable for interactive oral practice scenarios.
New prospects usually get a reply within 1 business day. We start with a 30-minute scoping call to align on scenario and metrics, then define a PoC integration plan based on business scale, assessment types, concurrency, and data-compliance requirements. The plan includes a shared technical channel, API integration support, and audio-quality tuning guidance before commercial quote and contract steps.
Public cloud uses tiered "usage × concurrency" pricing — unit price drops automatically at higher tiers without re-contracting. Concurrency packs are available for traffic peaks. Sales provides exact tier rates based on volume and scenario.