Pinecone Query

Pinecone Query Skill

Search for records in Pinecone integrated indexes using natural language text queries via the Pinecone MCP server.

<<clarify_style>>

What is this skill for?

This skill provides a simple way to query integrated indexes (indexes with built-in Pinecone embedding models) using text queries. The MCP server automatically converts your text into embeddings and searches the index.

Prerequisites

Required:

1. ✅ Pinecone MCP server must be configured - Check if MCP tools are available
2. ✅ PINECONE_API_KEY environment variable must be set - Get a free API key at https://app.pinecone.io/?sessionType=signup (opens in a new tab)
3. ✅ Index must be an integrated index - Uses Pinecone embedding models (e.g., multilingual-e5-large, llama-text-embed-v2, pinecone-sparse-english-v0)

When NOT to use this skill

Use the pinecone-cli skill instead if:

• ❌ Your index is a standard index (no integrated embedding model)
• ❌ You need to query with custom vector values (not text)
• ❌ You need advanced vector operations (fetch by ID, list vectors, bulk operations)
• ❌ Your index uses third-party embedding models (OpenAI, HuggingFace, Cohere)

MCP Limitation: The Pinecone MCP currently only supports integrated indexes. For all other use cases, use the pinecone-cli skill.

How it works

Utilize Pinecone MCP’s search-records tool to search for records within a specified Pinecone integrated index using a text query.

Workflow

IMPORTANT: Before proceeding, verify the Pinecone MCP tools are available. If MCP tools are not accessible:

• Inform the user that the Pinecone MCP server needs to be configured
• Check if PINECONE_API_KEY environment variable is set
• Direct them to the MCP setup documentation or the pinecone-help skill

1. Parse the user’s input for:

   • query (required): The text to search for.
   • index (required): The name of the Pinecone index to search.
   • namespace (optional): The namespace within the index.
   • reranker (optional): The reranking model to use for improved relevance.

2. If the user omits required arguments:

   • If only the index name is provided, use the describe-index tool to retrieve available namespaces and ask the user to choose.
   • If only a query is provided, use list-indexes to get available indexes, ask the user to pick one, then use describe-index for namespaces if needed.

3. Call the search-records tool with the gathered arguments to perform the search.

4. Format and display the returned results in a clear, readable table including field highlights (such as ID, score, and relevant metadata).

---

Troubleshooting

PINECONE_API_KEY is required. Get a free key at https://app.pinecone.io/?sessionType=signup (opens in a new tab)

If you get an access error, the key is likely missing. Ask the user to set it and restart their IDE or agent session: <<api_key_setup>>

IMPORTANT At the moment, the pinecone-query skill can only be used with integrated indexes, which use hosted Pinecone embedding models to embed and search for data. If a user attempts to query an index that uses a third party API model such as OpenAI, or HuggingFace embedding models, remind them that this capability is not available yet with the Pinecone MCP server.

• If required arguments are missing, prompt the user to supply them, using Pinecone MCP tools as needed (e.g., list-indexes, describe-index).
• Guide the user interactively through argument selection until the search can be completed.
• If an invalid value is provided for any argument (e.g., nonexistent index or namespace), surface the error and suggest valid options.

Tools Reference

• search-records: Search records in a given index with optional metadata filtering and reranking.
• list-indexes: List all available Pinecone indexes.
• describe-index: Get index configuration and namespaces.
• describe-index-stats: Get stats including record counts and namespaces.
• rerank-documents: Rerank returned documents using a specified reranking model.
• Ask the user interactively to clarify missing information when needed.

---