MCP
Comment moderation and intelligence for articles: bot and spam detection, multilingual sentiment analysis and constructive comment generation.
Installation
Les étapes d'installation sont dans la documentation ci-dessous. Documentation
Documentation
Tirée du README du projet sur GitHub, actualisée chaque semaine. Les liens ouvrent GitHub.
Sommaire
SentiSift MCP Server
Official Model Context Protocol (MCP) server for SentiSift. Lets MCP-compatible AI hosts (Claude Desktop, Cursor, VS Code, Continue, and others) call the SentiSift comment-moderation and intelligence API as a native tool.
What this unlocks
Once installed, your AI assistant can:
- Analyze comments directly in conversation: "Run these 50 comments through SentiSift and show me the bots and the sentiment distribution."
- Check your account state: "What's my SentiSift balance and when does my subscription renew?"
- Retrieve historical results: "Show me the sentiment history for yesterday's article on example.com/breaking-news."
- Self-diagnose service status: "Is the SentiSift API ready, or are models still loading?"
No code required. The assistant calls the tools through MCP and reports back.
Tools exposed
| Tool | Description |
|---|---|
analyze_comments | Submit a batch of comments plus an article URL, receive bot/spam flags, sentiment labels, commercial-content flags, moderation summary, and (on Pro/Enterprise) Intelligence + interleaved Influence comments. |
get_balance | Current comment balance, tier, feature flags, active grants, subscription state, Influence stats. |
get_health | Whether the SentiSift API is ready or still loading models. |
get_article_results | Historical scored comments for an article URL (Pro/Enterprise retain full history; Free/Starter return buffer state only). |
Install
You need Python 3.10+ and a SentiSift API key. Free keys (1,000 comments, no credit card) at sentisift.com/pricing.
Recommended: uvx (zero install, always latest)
uv runs published Python apps without persistent install. This is the install-free path.
# One-time: install uv (mac/linux)
curl -LsSf https://astral.sh/uv/install.sh | sh
# Verify it works (Ctrl-C after a few seconds; the server is meant to be
# launched by your AI host, not run in a terminal).
SENTISIFT_API_KEY=sk_sentisift_... uvx sentisift-mcp
Alternative: pip install
pip install sentisift-mcp
sentisift-mcp # run manually to verify
Host configuration
Each AI host has its own config file. Pick yours:
Claude Desktop
Edit the config file:
- macOS:
~/Library/Application Support/Claude/claude_desktop_config.json - Windows:
%APPDATA%\Claude\claude_desktop_config.json
Add the SentiSift entry to mcpServers:
{
"mcpServers": {
"sentisift": {
"command": "uvx",
"args": ["sentisift-mcp"],
"env": {
"SENTISIFT_API_KEY": "sk_sentisift_your_key_here"
}
}
}
}
Restart Claude Desktop. You should see a new tool indicator in the bottom-left corner of a new chat. Try: "Use SentiSift to check my balance."
Cursor
Cursor supports MCP via the Settings → MCP Servers panel. Add:
- Name:
sentisift - Command:
uvx - Arguments:
sentisift-mcp - Environment variables:
SENTISIFT_API_KEY=sk_sentisift_your_key_here
Or edit ~/.cursor/mcp.json directly with the same structure as the Claude Desktop example above.
VS Code (with the MCP extension)
Edit your user or workspace settings.json:
{
"mcp.servers": {
"sentisift": {
"command": "uvx",
"args": ["sentisift-mcp"],
"env": {
"SENTISIFT_API_KEY": "sk_sentisift_your_key_here"
}
}
}
}
Continue.dev
In ~/.continue/config.json:
{
"mcpServers": [
{
"name": "sentisift",
"command": "uvx",
"args": ["sentisift-mcp"],
"env": {
"SENTISIFT_API_KEY": "sk_sentisift_your_key_here"
}
}
]
}
Any other MCP-compatible host
The server speaks MCP over stdio. Point your host at uvx sentisift-mcp (or sentisift-mcp if installed via pip) and set SENTISIFT_API_KEY in its environment.
How to talk to the assistant
Once configured, natural-language prompts work. Some examples:
"Check my SentiSift balance."
"Run these 20 comments through SentiSift for the article at https://example.com/breaking-story. Here are the comments:" (paste them)
"Is the SentiSift API ready right now, or is it still loading?"
"Fetch the processed comments for https://example.com/article-42 from SentiSift."
The assistant chooses the right tool, passes the right arguments, and summarizes the results. If a call fails, the error response includes a deep link to the relevant docs section so the assistant can self-correct.
Privacy
- The server sends a
User-Agentheader identifying it assentisift-mcp/<version>so we can distinguish MCP traffic from direct SDK traffic in our usage logs. No other telemetry. - The server runs locally on your machine. It only calls out to
api.sentisift.com. - Your API key stays on your machine (in the host's env config). It is never logged to disk by this server.
Source
Source and issue tracker: gitlab.com/pickel-fintech/sentisift-sdks.
Links
- SentiSift API docs
- OpenAPI spec
- Python SDK (this MCP server wraps it)
- Node SDK
- Changelog
- Support
License
MIT — see LICENSE. The SentiSift API service that this client calls is governed separately by sentisift.com/terms.html; use of the API requires an API key from sentisift.com/pricing.
Questions fréquentes
Qu'est-ce que MCP ?
Comment moderation and intelligence for articles: bot and spam detection, multilingual sentiment analysis and constructive comment generation.
Comment installer MCP ?
Suivez les étapes d'installation de sa documentation, affichée sur cette page et dans son dépôt.
MCP est-il open source ?
Oui. Son code est sur GitHub sous licence MIT.
Qui maintient MCP ?
Il est publié par pickelfintech. AI.tn le référence sans le maintenir ni le recommander.
Issu des listes communautaires awesome-mcp-servers et du dépôt GitHub du projet. Être listé n'est pas une recommandation : vérifiez le code et les permissions d'un serveur avant de lui donner accès à vos données.