MCP-RAG Server: Use Cases & Combinations

Quick Summary

The MCP-RAG TypeScript server provides three main tools:

1. ML FAQ Retrieval - Search the vector database for ML concepts
2. Add Document to FAQ - Expand your knowledge base with new information
3. Web Search via Scrapeless - Find information from the web

These tools can be used individually or combined in powerful workflows for knowledge management and retrieval.

Note: The tool names (like machine-learning-faq-retrieval) are configurable and don't necessarily match your Qdrant collection names. By default, this tool queries the "ml_faq_collection" in Qdrant, but you can customize both the tool name and the target collection in your configuration.

1. Machine Learning FAQ Retrieval (machine-learning-faq-retrieval)

Use Cases:

• Find information about specific ML concepts
• Search for techniques to solve ML problems
• Explore methods in particular ML domains
• Get quick answers about known ML topics

Example Prompts:

Use machine-learning-faq-retrieval to find information about overfitting

Can you search the ML FAQ for information about neural networks? Use machine-learning-faq-retrieval

Use machine-learning-faq-retrieval to find best practices for feature selection

2. Adding Documents to Knowledge Base (add-document-to-faq)

Use Cases:

• Enrich your knowledge base with new ML concepts
• Save useful information for future reference
• Create a personalized knowledge repository
• Document specific techniques or methods

Example Prompts:

Use add-document-to-faq to add this information:
Text: "Random forests are ensemble learning methods that construct multiple decision trees during training and output the class that is the mode of the classes or mean prediction of the individual trees."
Category: "Ensemble Methods"
Source: "Introduction to ML"
Tags: ["random forests", "ensemble learning", "decision trees"]

Add this to the ML FAQ using add-document-to-faq:
Text: "Transfer learning is a machine learning technique where knowledge gained during training in one type of problem is used to improve generalization in another type of problem."
Category: "Deep Learning"
Source: "Advanced ML Course"
Tags: ["transfer learning", "pre-trained models", "fine-tuning"]

3. Web Search via Scrapeless (scrapeless-web-search)

Use Cases:

• Search for recent information not in your ML database
• Explore topics outside the ML domain
• Get up-to-date data on emerging technologies
• Find resources, tutorials, or articles on a topic

Example Prompts:

Use scrapeless-web-search to find recent advancements in quantum machine learning

I need information about the latest TensorFlow release. Use scrapeless-web-search

Use scrapeless-web-search with country=fr and language=fr to find French resources on deep learning

4. Using REPL for Advanced Operations

Use Cases:

• Manually add documents when the standard tool isn't available
• Run complex queries on the database
• Create or modify collections
• Advanced manipulation of embeddings or data

Example Prompt:

Use the repl tool to create a new collection in Qdrant called "deep_learning_concepts"

5. Combined Workflows

Search & Enrich:

First, use scrapeless-web-search to find information about "capsule networks". Then, use add-document-to-faq to add the most important findings to our ML knowledge base.

Verify Before Adding:

First, use machine-learning-faq-retrieval to check if we already have information about "self-supervised learning". If not, I'll provide content to add to our knowledge base.

Building a Thematic Knowledge Base:

I want to build a comprehensive knowledge base about reinforcement learning. Let's first use scrapeless-web-search to find key concepts, then systematically add each concept to our ML FAQ using add-document-to-faq.

Knowledge Gap Identification:

Let's compare what we know internally versus what's available online. Use machine-learning-faq-retrieval for "graph neural networks", then use scrapeless-web-search for the same query, and identify what we should add to our knowledge base.

Regular Knowledge Base Updates:

Let's implement a routine to keep our ML FAQ current. Use scrapeless-web-search for "latest machine learning techniques 2025", then use add-document-to-faq to add any new techniques we don't already have in our database.

Answering Complex Questions:

I need a comprehensive answer about hyperparameter tuning. First use machine-learning-faq-retrieval to check our knowledge base, then augment with scrapeless-web-search for the latest methods.

6. System Enhancement Use Cases

Quality Check:

Use machine-learning-faq-retrieval with several key ML terms like "neural networks", "reinforcement learning", and "clustering" to test the coverage of our knowledge base.

Progressive Extension:

Let's improve our ML FAQ by adding information about XGBoost. First, use scrapeless-web-search to gather information, then add a comprehensive document using add-document-to-faq.

Information Updates:

First, let's search for our current information about "transformers" using machine-learning-faq-retrieval, then use scrapeless-web-search to find the latest developments, and finally update our knowledge base using add-document-to-faq.

7. Customizing for Different Collections

If you want to work with multiple Qdrant collections, you can define additional MCP tools in your server.ts file:

// Example of adding a new tool for a different collection
server.tool(
  "deep-learning-retrieval",  // New tool name
  "Retrieve information from the deep learning knowledge base.",
  {
    query: z.string().describe("The search query")
  },
  async (params) => {
    // Create a Qdrant client configured for a different collection
    const dlQdrantClient = new QdrantClient({
      ...QDRANT_CONFIG,
      collection: "deep_learning_collection"  // Different collection name
    });
    
    const result = await dlQdrantClient.searchSimilar({ query: params.query });
    
    return {
      content: [{
        type: "text",
        text: JSON.stringify(result, null, 2)
      }]
    };
  }
);

This way, you can have multiple specialized knowledge bases in your Qdrant database.

Important Notes:

1. For the add-document-to-faq tool, always explicitly mention the tool name
2. For scrapeless-web-search, ensure your API key is correctly configured
3. If using REPL as an alternative, adapt the code to your specific needs
4. For maximum efficiency, combine different tools in a logical sequence
5. Always verify the information before adding it to your knowledge base
6. Tool names can be customized in your code and don't need to match your Qdrant collection names

These use cases will help you fully leverage your MCP-RAG server capabilities and progressively build a rich and useful knowledge base.