← Back to directory
C

Chinese Sensitive Words MCP Server

Community
Chinese sensitive/prohibited word detection MCP Server for Xiaohongshu, Douyin, Kuaishou and Bilibili
GitHub source repository ↗
★ 113 Stars Category · Other Popular
51FMRS · D

A Chinese content-compliance focused MCP server featuring multi-platform dictionaries (Xiaohongshu, Douyin, Kuaishou, Bilibili), four-level risk classification with replacement suggestions, plus homophone, character-skipping, phone number/URL detection and NER filtering. Free usage is 100 requests/day, unlimited with a TOKEN. It suits Chinese content creators and operations teams doing pre-publish checks, but its output is not a substitute for official platform moderation.

Reliability
8/20
Security and permissions
8/20
Maintenance
11/20
Documentation
12/20
Setup experience
12/20
Read the FMRS scoring method →

This is an MCP server for detecting sensitive and prohibited words in Chinese text, with dedicated dictionaries for Xiaohongshu, Douyin, Kuaishou and Bilibili. Its massive dictionary is updated daily and covers politics, pornography, violence, gambling, drugs, advertising-law superlatives and medical-efficacy claims. Results are classified by risk level — high (account ban), medium (throttling), low (recommend modification) and tip — and come with safe replacement suggestions. It also supports homophone variant detection, character-skipping detection, phone number/URL detection, and NER-based filtering to reduce false positives on place names, person names and organization names.

Tools

check_sensitive_words
Detect sensitive/prohibited words in text; useful for checking copy, product descriptions and livestream scripts for compliance
get_word_suggestions
Get safe replacement suggestions for a sensitive word marked by the detector

Setup

Add the following to your Claude Desktop / Cursor / Windsurf config file: {"mcpServers":{"chinese-sensitive-words":{"command":"npx","args":["-y","chinese-sensitive-words-mcp"]}}}. For Claude Code run: claude mcp add chinese-sensitive-words -- npx -y chinese-sensitive-words-mcp. For OpenClaw run: npx mcporter config add chinese-sensitive-words-mcp --stdio "npx -y chinese-sensitive-words-mcp". To remove the usage limit, set WORDSCHECK_ACCESS_TOKEN in env; to use a custom service address, also set WORDSCHECK_API_BASE.

claude_desktop_config.json
{
  "mcpServers": {
    "chinese-sensitive-words": {
      "command": "npx",
      "args": ["-y", "chinese-sensitive-words-mcp"]
    }
  }
}

Fit and risk

Best for

  • Creators and operators publishing Chinese content on Xiaohongshu, Douyin, Kuaishou and Bilibili
  • Brands, e-commerce and livestream teams that need pre-publish compliance checks on Chinese copy
  • Users who want sensitive-word detection inside Claude Desktop, Cursor, Windsurf, Claude Code or OpenClaw

Not for

  • Scenarios requiring moderation of non-Chinese content
  • Users expecting a fully offline local dictionary (the npm package calls a remote API, which can be pointed elsewhere via WORDSCHECK_API_BASE and WORDSCHECK_ACCESS_TOKEN)
  • Users who treat its output as an official platform moderation verdict

Required permissions

  • Execute the chinese-sensitive-words-mcp package via npx
  • Optionally pass WORDSCHECK_ACCESS_TOKEN as an environment variable to lift the free quota
  • Optionally set WORDSCHECK_API_BASE as an environment variable to point to a custom service address

Risks and side effects

  • Without a TOKEN the free quota is 100 requests/day and usage is limited beyond that
  • Detection only covers its own dictionary, so both misses and false positives are possible; it does not replace official platform moderation
  • When a custom WORDSCHECK_API_BASE is configured, your copy is sent to that address, so you must verify its trustworthiness
  • Judgements about sensitive or prohibited words change over time; although the dictionary is updated daily it may lag behind platform rule changes

Troubleshooting

  1. First confirm npx can fetch and run chinese-sensitive-words-mcp
  2. If you hit a quota limit, check whether WORDSCHECK_ACCESS_TOKEN is configured
  3. If the service cannot be reached, check whether WORDSCHECK_API_BASE is set correctly
  4. If you see false positives on person, place or organization names, rely on the NER-based filtering reflected in the results
  5. If the client does not list the tools, make sure the config is under the correct mcpServers key and restart the client

Use cases

Check Xiaohongshu copy for prohibited words and advertising superlatives
Check Douyin livestream scripts and traffic-diversion wording for compliance
Check product descriptions for medical-efficacy claims and false advertising
Get safe replacement suggestions for flagged sensitive words
Detect homophone variants (e.g. “薇信” for WeChat) and character-skipping obfuscation (e.g. “加 微 信”)
Detect phone numbers and external links in copy

Supported clients

Claude DesktopFull support
CursorFull support
WindsurfFull support
Claude CodeFull support
OpenClawFull support