• SCHEMA
  • SEARCH
  • DATA
  • HUMAN

How to debug RAG retrieval with your AI client

Connect LanceDB to your AI client to inspect your vector data directly. You can then run targeted searches to see exactly what your RAG system sees.

Ask AI about this page

Short answer

Why is my RAG system retrieving the wrong documents?

Your AI can pull your table schemas and run similarity searches to check if your embeddings or metadata are causing the issue. This lets you verify if the right context is actually available in your vector store.

Debugging outcomes.

Where every issue lands.

What you get after running these checks.

SCHEMA

Verify table structure

The AI inspects your LanceDB tables to ensure your metadata and vector columns match what your application expects.

SEARCH

Test similarity logic

Run specific queries against your vectors to see which documents come up and why they match your input.

DATA

Spot missing context

Identify if the documents you need are actually indexed or if they were never added to the database.

HUMAN

Fix the embeddings

If the search results look wrong, you'll need to manually adjust your embedding model or chunking strategy.

The workflow

What your AI does when <the thing arrives>.

The AI acts as a bridge between your chat interface and your LanceDB instance.

  1. Inspect the database

    The AI looks at your LanceDB tables to understand the schema, including vector dimensions and metadata fields.

    get_table

  2. Run similarity tests

    The AI performs searches using your query to find the closest vector matches in the database.

    vector_search

  3. Compare results

    The AI evaluates the retrieved documents against your original question to see if the context is relevant.

    vector_search

  4. Report discrepancies

    The AI tells you if the schema is broken or if the similarity scores are too low for a good match.

    get_tablevector_search

Try it

Copy these to start.

Use these prompts to begin your debugging session.

Starting points

Replace the table names and query strings with your actual LanceDB details.

LanceDB Connector

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

Connector linkhttps://edge.vinkius.com/vk_preview_85lSO635ajE8lXHCZKWQ7UC6VHfybBdy20nU97wm/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 — LanceDB capabilities are ready to use.
Configuration · claude_desktop_config.jsonCopy
{
  "mcpServers": {
    "lancedb-serverless-vector-db-mcp": {
      "url": "https://edge.vinkius.com/vk_preview_85lSO635ajE8lXHCZKWQ7UC6VHfybBdy20nU97wm/mcp"
    }
  }
}
Copy into chat04

  • Show me the schema for my 'documents' table in LanceDB.

  • Run a similarity search in the 'embeddings' table for the phrase 'how to fix a leak'.

  • Check the 'user_data' table and find the top 3 matches for 'account settings'.

  • Why isn't my query about 'pricing' returning any results from the 'product_info' table?

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

Start here

Connect LanceDB once, then ask.

Just link your account once. Your credentials stay encrypted, and you can immediately start asking your AI about your data.

Connect LanceDB to your AI

FAQ

How this task behaves.

  • 01

    Can the AI delete my data?

    No. The AI only has permission to read schemas and perform searches.

  • 02

    Can the AI update my vector embeddings?

    No. The AI can only inspect and search existing data; it cannot write or modify records.

  • 03

    Does the AI see all my tables?

    The AI can see any table you explicitly tell it to inspect using the provided tools.

  • 04

    Can I use this with any vector database?

    No, this specific MCP is built exclusively for LanceDB.

  • 05

    Will the AI suggest new chunking strategies?

    The AI can analyze why current chunks are failing, but you must implement any changes to your code yourself.

  • More questions about LanceDB? See everything the LanceDB Connector can do

Connect LanceDB to Claude, Cursor, ChatGPT & more