semantic-searchLABSCO SUMMARY
An agent can put things into your vector store as well as search it.
The write side is what makes it more than a search tool: hand it a document and it chunks, embeds and upserts, so the same conversation that failed to find an answer can add the source that would have had it.
What it is
A read-and-write client for one Pinecone index. It embeds and searches, reads a document back out, lists what the index holds, and processes a document into chunks and upserts them.
What you get
- Semantic search across the records in the index
- A document read back out of the index
- Every document in the index, listed
- A document chunked, embedded and upserted
- Index statistics: record count, dimensions and namespaces
Requirements
Python with the mcp-pinecone package, a Pinecone API key, and an index to point at. Written for Claude Desktop, and works from any client that can launch a local process.
Setup effort
One command plus a key — uv pip install mcp-pinecone, then supply credentials
USE CASES
01
Finding the passage that answers a question
Search the index by meaning, then read the document behind the hit rather than working from the snippet.
02
Adding a document the index was missing
Process it into chunks and upsert them, then check the record count moved the way you expected.
03
Checking what is actually in the index
List the documents and read the index statistics — records, dimensions, namespaces — before blaming the query for a bad result.
04
Reading the full source behind a search hit
Take the id off the search result and pull the whole document, so the answer can quote rather than paraphrase.
5 tools
Listed in project docs. Your client may show a different list.
read-documentlist-documentspinecone-statsprocess-documentBEFORE YOU INSTALL
What to know first
Tools you get
5 tools named in the project's documentation.
Requires
uvx on PATH
Where it runs
Both documented setups run the server on your machine.
INSTALLATION
Connect your client
Server setup
TERMINAL
uv pip install mcp-pineconeMCP Clients
Claude Desktop — uvx
CLAUDE DESKTOP — UVX · JSON
{
"mcpServers": {
"mcp-pinecone": {
"command": "uvx",
"args": [
"--index-name",
"<YOUR_INDEX_NAME>",
"--api-key",
"<YOUR_SECRET_API_KEY>",
"mcp-pinecone"
]
}
}
}Claude Desktop — local dev
CLAUDE DESKTOP — LOCAL DEV · JSON
{
"mcpServers": {
"mcp-pinecone": {
"command": "uv",
"args": [
"--directory",
"<YOUR_PROJECT_DIR>",
"run",
"mcp-pinecone"
]
}
}
}
