analyze_context
Analyze text with context awareness - considers domain-specific whitelists
LABSCO SUMMARY
The parts a word list cannot do: obfuscation, context, and a pattern across a user's messages.
analyze_context applies a domain whitelist so clinical or gaming vocabulary is not treated as abuse, detect_obfuscation catches leetspeak and homoglyph tricks, and explain_match gives a moderator a reason rather than a verdict. track_user_message is the one that changes moderation practice: it turns individual decisions into a profile, which is how a single outburst stays distinguishable from a pattern worth acting on.
What it is
A moderation server built on the glin-profanity library. It flags and censors text across 24 languages, sees through leetspeak and Unicode substitution, and keeps a risk profile per user id.
What you get
- Text checked with the matched words, their severity and their positions in the string — check_profanity
- Matches replaced with a character you choose, optionally keeping the first letter visible — censor_text
- Domain-aware analysis, where a whitelist for that domain stops ordinary specialist vocabulary being flagged — analyze_context
- Up to 100 texts checked in one call, returning only the flagged ones if that is all you need — batch_check
- A safety score from 0 to 100 with a recommended action, at the strictness you set — validate_content
- Obfuscation identified as such: leetspeak and Unicode homoglyphs — detect_obfuscation
- The reasoning behind one flagged word, and cleaner rewordings in a tone you pick — explain_match, suggest_alternatives
- Statistics across a collection of up to 500 texts, with word frequency — analyze_corpus
- The same text compared across strictness levels, and a regex pattern generated for a word you want to catch yourself — compare_strictness, create_regex_pattern
- Per-user moderation: messages tracked against a user id, profiles and risk read back, high-risk users listed for review, and a profile reset after an appeal — track_user_message, get_user_profile, get_high_risk_users, reset_user_profile
- A streaming path for chat: one message or a batch processed against a queue, with throughput statistics — stream_check, stream_batch, get_stream_stats
- The supported languages listed, and per-call options for custom words, ignore words, case sensitivity and word boundaries — get_supported_languages
Requirements
No account and no key. npx -y glin-profanity-mcp runs it over stdio, or install glin-profanity-mcp globally; the package also ships an HTTP entry point, glin-profanity-mcp-http, for clients that need one. Detection options travel with each call — languages, custom words, ignore words, case sensitivity, word boundaries, leetspeak detection, Unicode normalisation — so there is no configuration file to prepare first.
Setup effort
One command — npx -y glin-profanity-mcp
USE CASES
01
Screening user text before it goes live
Check it, then take the safety score and its recommended action rather than deciding from the raw match list.
02
Publishing a comment with the words masked instead of rejecting it
Censor the matches with your own replacement character, keeping first letters if that is your house style.
03
Cutting false positives in a community with its own vocabulary
Analyse with the domain set, and ask why anything still flagged was flagged.
04
Moderating a live chat without adding lag
Push messages through the streaming queue and watch the throughput statistics as volume rises.
05
Deciding whether a user needs a human review
Track their messages, read the risk profile that builds up, and pull the list of the highest-risk accounts.
06
Auditing a backlog of comments you inherited
Run them in batches, then take word-frequency statistics over the whole corpus to see what you are dealing with.
19 tools
Counted once with only startup setup. Other versions, modes or optional credentials may expose a different list.
analyze_corpus
Analyze a collection of texts for profanity statistics (max 500)
batch_check
Check multiple texts for profanity in a single operation (max 100)
censor_text
Censor profanity in text by replacing matched words with a replacement character
check_profanity
Check text for profanity and get detailed results including matched words, severity, and positions
compare_strictness
Compare detection results across different strictness levels
create_regex_pattern
Generate regex patterns for custom profanity detection
detect_obfuscation
Detect text obfuscation techniques like leetspeak and Unicode homoglyphs
explain_match
Explain why a word was flagged with detailed reasoning
get_high_risk_users
Get list of users with high risk scores for review
get_stream_stats
Get real-time streaming statistics and performance metrics
get_supported_languages
Get the list of all 24 supported languages
get_user_profile
Get a user's moderation profile and risk assessment
reset_user_profile
Reset a user's moderation profile (use after appeals or timeouts)
stream_batch
Process multiple messages in real-time batch mode - for high-throughput scenarios
stream_check
Add message to real-time processing queue and get instant result - optimized for chat streams
suggest_alternatives
Suggest clean alternatives for profane content
track_user_message
Track a user message and update their risk profile - enables pattern detection across messages
validate_content
Validate content with a safety score (0-100) and action recommendation
BEFORE YOU INSTALL
What to know first
Tools you get
19 tools when we connected; the project documents 24. The 5 AI guardrail tools (check_prompt_injection, scan_secrets, scan_pii, redact_pii, restore_pii) are documented but absent from the current npm build.
Requires
npx on PATH
Where it runs
All 3 documented clients run the server on your machine.
INSTALLATION
Connect your client
Server setup
TERMINAL
npx -y glin-profanity-mcpMCP Clients
Claude Desktop, Cursor, Windsurf
JSON
{
"mcpServers": {
"glin-profanity": {
"command": "npx",
"args": [
"-y",
"glin-profanity-mcp"
]
}
}
}
