Sentor MCP Server

Entity-based sentiment analysis for Claude, Cursor, Windsurf, and any MCP-compatible AI assistant.

Sentor is an entity-based sentiment analysis platform powered by fine-tuned BERT models. This MCP server exposes Sentor's ML APIs as tools your AI assistant can call directly — score sentiment toward specific entities in text, cluster documents by topic, and generate topic labels, all from a single natural-language prompt.

---

Table of Contents

• What It Does
• Requirements
• Quick Start
  • Claude Desktop
  • Cursor / Windsurf
  • Claude.ai Web (Remote MCP)
• Tools Reference
• Usage Examples
• Rate Limits
• Remote Deployment
• Links

---

🎯 What It Does

Once connected, your AI assistant gains four tools:

Tool	What it does
analyze_sentiment	Score sentiment toward named entities (brands, products, features, people) in one or more documents. Returns per-document and per-sentence breakdowns.
cluster_documents	Group 5+ documents into thematic clusters using BERTopic + HDBSCAN. Automatically discovers the number of clusters.
name_topic	Generate a 3–5 word descriptive label for each cluster using an LLM (e.g. "Shipping Delay Complaints").
health_check	Verify the Sentor API is reachable and ML models are loaded.

Example prompt after setup:

"Analyse these 50 customer reviews for sentiment toward our checkout flow and delivery speed. Then cluster them by topic and name each cluster."

---

📋 Requirements

• Python 3.10+
• A Sentor API key — get one free at dashboard.sentor.app

---

🚀 Quick Start

Claude Desktop

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "sentor": {
      "command": "uvx",
      "args": ["sentor-mcp"],
      "env": {
        "SENTOR_API_KEY": "your_api_key_here"
      }
    }
  }
}

Restart Claude Desktop. A hammer icon appears in the tool selector — Sentor is ready.

No uvx? Install it with pip install uv, or use sentor-mcp directly after pip install sentor-mcp.

---

Cursor / Windsurf

Add to .cursor/mcp.json (project-level) or ~/.cursor/mcp.json (global):

{
  "mcpServers": {
    "sentor": {
      "command": "uvx",
      "args": ["sentor-mcp"],
      "env": {
        "SENTOR_API_KEY": "your_api_key_here"
      }
    }
  }
}

---

Claude.ai Web (Remote MCP)

Run the HTTP server and connect by URL:

docker run -e SENTOR_API_KEY=your_api_key -p 8080:8080 ghcr.io/nikx-tech/sentor-mcp:latest

Then in Claude.ai → Settings → Integrations → Add MCP Server:

http://your-server:8080/sse

---

🔧 Tools Reference

analyze_sentiment(docs, language="en")

Analyse entity-level sentiment in one or more documents.

docs = [
    {
        "doc_id": "review-1",
        "doc": "The delivery was fast but the packaging was completely crushed.",
        "entities": ["delivery", "packaging"]
    }
]
# Returns: predicted_label, probabilities, per-sentence details

Supported languages: en (English), nl (Dutch)

---

cluster_documents(documents, language="en")

Group documents into thematic clusters. Requires at least 5 documents.

documents = [
    {"doc_id": "r1", "text": "Great product quality, very happy.", "entities": ["product"]},
    # ... at least 5 documents
]
# Returns: clusters with cluster_id, document_count, documents, top_words
# Cluster -1 = outliers that did not fit any topic

---

name_topic(cluster_id, documents, top_words, entities, language="en")

Generate a short label for a cluster. Pass data directly from cluster_documents output.

name_topic(
    cluster_id=0,
    documents=cluster["documents"],
    top_words=cluster["top_words"],
    entities=["BrandName"],  # exclude your brand from the label
    language="en"
)
# Returns: { "topic_name": "Shipping Delay Complaints", "generation_method": "LLM" }

---

health_check()

# Returns: { "status": "healthy", "version": "1.0.0", "llm_status": "available" }

---

💬 Usage Examples

Single document:

"Use Sentor to analyse the sentiment of this review toward Apple and iPhone: [paste text]"

Batch analysis:

"I have 100 customer reviews. Use Sentor to score sentiment toward 'delivery' and 'support' in each one, then tell me the ratio of positive to negative."

Full pipeline:

"Use Sentor to: 1) analyse sentiment in these 200 reviews for 'product quality' and 'price', 2) cluster them by topic, 3) name each cluster, 4) summarise the findings."

Competitive analysis:

"Analyse these tweets for sentiment toward Apple, Samsung, and Google separately using Sentor, then compare the results."

---

📊 Rate Limits

Plan	Per Minute	Per Day	Per Month
Free	5	100	1,000
Starter	60	1,000	10,000
Growth	200	3,000	30,000
Business	500	10,000	100,000
Enterprise	Custom	Custom	Custom

View full pricing →

---

🐳 Remote Deployment

Run as a hosted HTTP/SSE server for AI tools that support remote MCP endpoints.

Docker:

docker build -t sentor-mcp .
docker run \
  -e SENTOR_API_KEY=your_key \
  -p 8080:8080 \
  sentor-mcp

The server exposes:

• GET /sse — SSE stream (MCP transport)
• POST /messages — message endpoint

Environment variables:

Variable	Default	Description
SENTOR_API_KEY	—	Required. Your Sentor API key.
SENTOR_BASE_URL	https://sentor.app/api	Override to point at a self-hosted Sentor instance.
PORT	8080	HTTP server port.

---

🔗 Links

• Sentor Dashboard — manage API keys, projects, and usage
• API Documentation — full REST API reference
• MCP Integration Guide — step-by-step setup
• PyPI Package — pip install sentor-mcp
• Support

---

Built by NIKX Technologies B.V.