Keyphrases MCP

🚀 Keyphrases-MCP

Empowering LLMs with authentic keyphrase Extraction

This Keyphrases MCP Server is a natural language interface crafted for agentic applications. Its core function is to extract keyphrases from the provided text. By integrating smoothly with MCP (Model Content Protocol) clients, it allows AI - driven workflows to extract keyphrases more precisely and relevantly through the BERT machine - learning model. It directly processes local files in permitted directories and saves context tokens for the LLM. Moreover, it guarantees secure document processing by only exposing the extracted keyphrases to the MCP client, not the original file content.

🚀 Quick Start

With this MCP Server, you can pose the following queries:

• "Extract 7 keyphrases from the file. [ABSOLUTE_FILE_PATH]"
• "Extract 3 keyphrases from the given file ignoring the stop words. Stop words: former, due, amount, [OTHER_STOP_WORDS]. File: [ABSOLUTE_FILE_PATH]"

Keyphrases are extremely useful as they enable users to quickly understand the main topics and themes of a document without having to read it in its entirety. They also support the following applications:

1. Serve as tags or metadata for documents, enhancing organization and discoverability in digital libraries.
2. Help identify emerging trends and sentiment from customer reviews, social media, or news articles.
3. Act as features or inputs for other tasks, such as text classification and clustering.

✨ Features

Reasoning for keyphrases - mcp

Autoregressive LLM models, like those in Claude or ChatGPT, process text sequentially. This not only restricts their ability to fully contextualize keyphrases across the whole document but also leads to context degradation as the input length grows, causing earlier keyphrases to receive less attention.

Bidirectional models, such as BERT, take into account both the left and right context and maintain more consistent attention throughout the sequence. Generally, they can extract existing keyphrases from texts more accurately and relevantly, especially when no domain - specific fine - tuning is applied.

However, as autoregressive models adopt longer context windows and techniques like input chunking, their performance in keyphrase extraction is improving, gradually closing the gap with BERT. Additionally, domain - specific fine - tuning can make autoregressive LLM models outperform the BERT solution.

This MCP server combines BERT for keyphrase extraction with an autoregressive LLM for text generation or refinement, enabling seamless text processing.

🔧 Technical Details

How it works

The server employs a KeyBERT framework for the multi - step extraction pipeline, which combines spaCy NLP preprocessing with BERT embeddings:

1. Candidate Generation: The KeyphraseCountVectorizer identifies meaningful keyphrase candidates using spaCy's en_core_web_trf model and filters out stop words.
2. Semantic Encoding: Both candidates and the document are embedded using the paraphrase - multilingual - MiniLM - L12 - v2 sentence transformer.
3. Relevance Ranking: KeyBERT calculates the cosine similarity between candidate keyphrase and document embeddings.
4. Diversity Selection: Maximal Marginal Relevance (MMR) ensures the selection of diverse, non - redundant keyphrases.
5. Final Output: The top N most relevant and diverse keyphrases are selected and sorted alphabetically.

For more details, refer to the configuration document.

📚 Documentation

• 🚀 Integration - Instructions on integrating the server with your MCP client or LLM.
• 🔧 Configuration - Information about environment variables and settings.
• 🛠️ MCP Tools - A list of all available tools.
• 🪵 Development and testing - A guide on how to contribute to the project.

📄 License

This project is licensed under the MIT License.

📞 Contact

For any questions or support, please use GitHub Issues.