- SEMANTIC
- DQL
- RELEVANCE
- HUMAN
Short answer
How can I search my unstructured data with my AI?
Your AI uses semantic search and DQL to scan watsonx Discovery collections for specific information. This lets you ask natural questions and get precise answers from your documents. Here is how the process works.
Where every insight lands.
The outcomes of your queries
Depending on how you phrase your request, your AI delivers different types of intelligence.
SEMANTIC
Natural language answers
The AI finds meaning behind your words rather than just matching keywords. You get direct answers based on the context of your documents.
DQL
Structured data hits
Use Discovery Query Language to filter for specific metadata or fields. This pulls exact records that meet your strict criteria.
RELEVANCE
Ranked document lists
The AI returns a list of the most relevant files or snippets. You can then pick the best source to verify the information.
HUMAN
Manual verification
The AI identifies the source documents for every answer. You decide if the extracted insight is accurate enough for your needs.
The workflow
What your AI does when a query arrives.
The AI handles the heavy lifting of scanning through massive datasets so you don't have to.
Interpret intent
Your agent looks at your question to decide if it needs a broad semantic search or a strict DQL filter.
query_discovery_contentScan collections
The AI reaches into your specific watsonx Discovery collections to find matching text or metadata.
query_discovery_contentExtract context
It pulls the relevant snippets and surrounding text to ensure the answer makes sense.
query_discovery_contentSynthesize response
The AI turns the raw data from the discovery tool into a plain English answer for you.
query_discovery_content
Try it
Copy these to start.
Use these prompts to test the connection immediately.
Starting points
Replace the bracketed text with your actual collection names or specific document topics.
watsonx Discovery Connector
You're all set. Choose your MCP client and follow the setup instructions.
https://edge.vinkius.com/vk_preview_MOvCoAeU4n0XLtoAj5AMd7HKRadaRJgAFtxeD6Bh/mcpClaude Desktop
Follow the steps below to connect in seconds.
- 1In Claude Desktop, open Settings → Connectors.
- 2Click “Add custom connector” and paste the connector link above as the remote MCP server URL.
- 3Click Add and start a new chat — watsonx Discovery capabilities are ready to use.
{
"mcpServers": {
"watsonx-discovery-mcp": {
"url": "https://edge.vinkius.com/vk_preview_MOvCoAeU4n0XLtoAj5AMd7HKRadaRJgAFtxeD6Bh/mcp"
}
}
}Search the [Collection Name] collection for any mention of recent contract updates.
Using DQL, find all documents in [Collection Name] where the status is 'active'.
What are the main takeaways from the technical manuals in my [Collection Name] collection?
Find the specific clause in the [Collection Name] regarding liability limits.
Claude
ChatGPT
Cursor
VS Code
Windsurf
Claude Code
JetBrains
Cline
Start here
Connect watsonx Discovery once, then ask.
Follow the link to authorize the connection. Your credentials stay encrypted, and you can start querying immediately.
Connect watsonx Discovery to your AIFAQ
How the search behaves
- 01
Can the AI delete documents in my Discovery collection?
No. The AI can only read and query content through the provided tools; it has no permission to modify or delete your data.
- 02
Does the AI search all my data or just specific collections?
It only searches the specific collections you point it toward in your prompt.
- 03
Can the AI create new documents in watsonx Discovery?
No. The tool is strictly for querying existing content.
- 04
How does it handle complex queries?
It uses DQL for structured filtering and semantic search for natural language questions.
- 05
Will the AI tell me which document it used for an answer?
Yes, it uses the data returned from the query to identify the source of the information.
More questions about watsonx Discovery? The Connector page answers them. See everything the watsonx Discovery Connector can do
Connect watsonx Discovery to Claude, Cursor, ChatGPT & more
