Mandarin audio, in your conversation

Chinese Text to Speech for Claude

Try the online Chinese reader →Connect the plugin

Turn Mandarin Chinese and mixed Chinese–English text into downloadable MP3 audio. Choose a male or female voice and a speaking speed from 0.5 to 2.0.

Independent service by Uranbek Anarbaev. Version 1.2.0 · Updated 9 October 2026. Free MCP quota; Claude host plan requirements may apply.

How does Chinese speech generation work?

1. 1. Send your textAsk Claude for Mandarin audio and choose a voice.
2. 2. Generate through MCPThe approved production connector checks quota and calls a speech provider.
3. 3. Download MP3Claude returns the recording link. Save it before expiry.

Listen to real voice samples

Both recordings say 你好, “Hello world!” at normal speed. They were generated by this production MCP. These intentionally public demos are permanent; your own generated links expire after 24 hours. Provider fallback may change the exact voice.

Example

Ask: “Read 你好，Hello world! slowly in a female voice.” The plugin returns an audio download link valid for 24 hours.

How to connect the Claude plugin

Production HTTP MCP: https://api-chinese-tts.hsk-tutor.com/mcp. Approve the bundled MCP when installing the plugin. If your Claude app only imports the skill, add this URL separately as a custom connector; availability depends on your plan and administrator settings.

Download plugin ZIP · v1.2.0Official MCP setup reference

In Claude Code, extract the plugin and run claude --plugin-dir /absolute/path/to/plugin. The same package uses production for testing and release, without a local speech server. Ask for a short recording first. Successful generation returns a real MP3 download link, not simply a written description of audio.

Mandarin pronunciation and mixed English audio

Start with a short sentence you understand. Listen while following the Chinese characters, then repeat it aloud. Slower speech helps you notice syllables and pauses; normal speed is useful for listening practice. Generated pronunciation is a study aid, not a guarantee of the correct reading for every name, abbreviation or character with multiple pronunciations.

The plugin accepts Chinese text with English words mixed in. A passage such as 你好，Hello world! can be submitted in one request. Voice options are female and male, with speeds between 0.5 and 2.0. The speech provider may vary, so a voice is not a permanently fixed speaker identity. Cantonese voices, offline synthesis and voice cloning are not offered.

Three requests to try

• “Read 你好，很高兴认识你。 in a female voice.”
• “Generate a male-voice MP3 for 今天我们学习 Chinese pronunciation.”
• “Read 请问，地铁站在哪里？ at speed 0.8 so I can repeat it.”

Claude returns the audio link after generation. Download it within 24 hours if you want to keep it. The plugin does not evaluate a recording of your own voice or score your pronunciation: it synthesizes the written text you supply.

Free quota

500 units per anonymous client ID per UTC day, within a shared 10,000-unit daily allowance. A Chinese character or English word is one unit. Numeric tokens count as one; punctuation is free. Failed synthesis consumes reserved units. Anonymous identifiers are not verified per-person accounts.

Need more? Use the website, with its separate limits.

Claude plugin or online reader?

Use Claude for a quick sentence while you are already chatting. Use the website to edit a passage, follow characters with pinyin, or prepare a longer reading exercise with document tools. Check the reader’s current allowance before generating long recordings. Neither route promises unlimited use. If the plugin’s shared daily quota is exhausted, the website may remain available under its separate limits.

The service works online and needs a connected, approved MCP. If tools are missing, check your Claude connector settings and administrator permissions. If a link has expired, it cannot be downloaded; generating again is a new request and uses quota. Avoid repeated retries when the speech service reports an error because failed generation still consumes reserved units.

Audio privacy and practical next steps

The text you supply is sent to external speech providers. Anyone holding a recording’s random download URL can access it during its 24-hour lifetime. Do not submit confidential information. Expired links stop working, but physical file cleanup can occur later. Review the dedicated MCP privacy policy before submitting sensitive material.

For listening practice, choose two or three sentences on a familiar topic. Listen once with the characters visible and once without looking. Repeat one sentence and compare the pauses. Keep your source passage next to a downloaded MP3 so the sound remains connected to meaning. Use the reader’s pinyin view when you need support, then return to the characters as you become more comfortable.

Open the Chinese reader and prepare your next recording →

Learn how to save Chinese speech as MP3 or practice reading with pinyin. Developed by Uranbek Anarbaev; independent of Anthropic. For connection or audio errors, see plugin support.

Frequently asked questions

For host-specific setup, consult Claude’s official MCP documentation. Compatibility and directory review are separate from the service’s generation tests.

Privacy · Terms · Support