• MATCHES
  • DOCUMENTS
  • MISSES
  • HUMAN

How to debug RAG retrieval with your AI agent

Connect DataStax Astra DB Vector once. After that, your agent can peek into your database to see the specific data hitting your prompt.

Ask AI about this page

Short answer

How can I debug RAG retrieval with my AI?

You can ask your agent to pull specific vector matches or individual JSON documents from Astra DB. This lets you verify if the retrieved context actually matches your query. You'll see the exact data being used.

Where your debug data lands.

The outcomes of your inspection.

Your agent uses these tools to pull evidence from your database.

MATCHES

Verify vector hits

The AI finds the closest vectors to a query. You get the specific chunks that the RAG pipeline is currently seeing.

DOCUMENTS

Inspect raw JSON

Your agent pulls the full JSON objects for specific records. This helps you check if metadata or content fields are formatted correctly.

MISSES

Identify retrieval gaps

By comparing what the AI finds against what you expected, you can spot where your embedding model or indexing is failing.

HUMAN

Fix the index

The AI cannot rewrite your database or change your embedding logic. You'll need to manually adjust your data based on what the agent finds.

The workflow

What your AI does when <the thing arrives>.

The agent acts as a bridge between your chat interface and your Astra DB instance.

  1. Find similar vectors

    Your agent searches the vector index to find the most relevant pieces of information for a given query.

    vector_search

  2. Pull specific records

    Once a relevant document is identified, the agent grabs the full record to show you the complete context.

    find_one_document

  3. List multiple entries

    The agent can pull a set of documents to help you compare multiple potential matches at once.

    find_documents

  4. Analyze the results

    The agent looks at the retrieved data and compares it to your original question to tell you if the retrieval was successful.

    vector_searchfind_one_document

Try it

Copy these to start.

Use these prompts to begin debugging your retrieval.

Starting points

Replace the bracketed text with your actual collection names or query strings.

DataStax Astra DB Connector

You're all set. Choose your MCP client and follow the setup instructions.

Connector linkhttps://edge.vinkius.com/vk_preview_AqDRfth2dYqeDTcgeoWzkACu9uC1WOaTh5GDBWN7/mcp

Claude Desktop

Follow the steps below to connect in seconds.

  1. 1In Claude Desktop, open Settings → Connectors.
  2. 2Click “Add custom connector” and paste the connector link above as the remote MCP server URL.
  3. 3Click Add and start a new chat — DataStax Astra DB capabilities are ready to use.
Configuration · claude_desktop_config.jsonCopy
{
  "mcpServers": {
    "datastax-astra-db-vector-mcp": {
      "url": "https://edge.vinkius.com/vk_preview_AqDRfth2dYqeDTcgeoWzkACu9uC1WOaTh5GDBWN7/mcp"
    }
  }
}
Copy into chat04

  • Run a vector search in the 'customer_support' collection for the query 'how do I reset my password' and show me the results.

  • Find the document with ID 'user_12345' in Astra DB and tell me if the metadata contains the correct region.

  • Search the 'product_catalog' collection for 'wireless headphones' and list the top 5 documents found.

  • Check the vector search results for 'shipping policy' and tell me if the retrieved text actually answers the question.

  • Claude
  • ChatGPT
  • Cursor
  • VS Code
  • Windsurf
  • Claude Code
  • JetBrains
  • Cline

Start here

Connect DataStax Astra DB Vector once, then ask.

Just provide your credentials through the secure connection. Once it's live, your agent can immediately start querying your database.

Connect DataStax Astra DB Vector to your AI

FAQ

How this task behaves.

  • 01

    Can the AI delete data from my Astra DB?

    No. The tools only allow the AI to search and find documents. It cannot delete or modify your data.

  • 02

    Can the AI update my vector embeddings?

    No. The AI can only read what is already in the database. You must handle embedding updates through your own pipeline.

  • 03

    Does the AI see my database credentials?

    No. Your credentials are encrypted and managed by Vinkius. Your agent only sees the data returned by the tools.

  • 04

    Can I use this to fix my RAG accuracy?

    The AI can show you why the accuracy is low by revealing what is being retrieved, but you have to perform the actual fixes in your code or data.

  • 05

    Can the AI perform complex aggregations?

    No. The AI is limited to the specific search and retrieval tools provided in the manifest.

  • More questions about DataStax Astra DB Vector? The Connector page answers them. See everything the DataStax Astra DB Vector Connector can do

Connect DataStax Astra DB to Claude, Cursor, ChatGPT & more