Read and write to a Pinecone index.
flowchart TB
subgraph Client["MCP Client (e.g., Claude Desktop)"]
UI[User Interface]
end
subgraph MCPServer["MCP Server (pinecone-mcp)"]
Server[Server Class]
subgraph Handlers["Request Handlers"]
ListRes[list_resources]
ReadRes[read_resource]
ListTools[list_tools]
CallTool[call_tool]
GetPrompt[get_prompt]
ListPrompts[list_prompts]
end
subgraph Tools["Implemented Tools"]
SemSearch[semantic-search]
ReadDoc[read-document]
ListDocs[list-documents]
PineconeStats[pinecone-stats]
ProcessDoc[process-document]
end
end
subgraph PineconeService["Pinecone Service"]
PC[Pinecone Client]
subgraph PineconeFunctions["Pinecone Operations"]
Search[search_records]
Upsert[upsert_records]
Fetch[fetch_records]
List[list_records]
Embed[generate_embeddings]
end
Index[(Pinecone Index)]
end
%% Connections
UI --> Server
Server --> Handlers
ListTools --> Tools
CallTool --> Tools
Tools --> PC
PC --> PineconeFunctions
PineconeFunctions --> Index
%% Data flow for semantic search
SemSearch --> Search
Search --> Embed
Embed --> Index
%% Data flow for document operations
UpsertDoc --> Upsert
ReadDoc --> Fetch
ListRes --> List
classDef primary fill:#2563eb,stroke:#1d4ed8,color:white
classDef secondary fill:#4b5563,stroke:#374151,color:white
classDef storage fill:#059669,stroke:#047857,color:white
class Server,PC primary
class Tools,Handlers secondary
class Index storage
The server implements the ability to read and write to a Pinecone index.
semantic-search
: Search for records in the Pinecone index.read-document
: Read a document from the Pinecone index.list-documents
: List all documents in the Pinecone index.pinecone-stats
: Get stats about the Pinecone index, including the number of records, dimensions, and namespaces.process-document
: Process a document into chunks and upsert them into the Pinecone index. This performs the overall steps of chunking, embedding, and upserting.
Note: embeddings are generated via Pinecone's inference API and chunking is done with a token-based chunker. Written by copying a lot from langchain and debugging with Claude.
To install Pinecone MCP Server for Claude Desktop automatically via Smithery:
npx -y @smithery/cli install mcp-pinecone --client claude
Recommend using uv to install the server locally for Claude.
uvx install mcp-pinecone
OR
uv pip install mcp-pinecone
Add your config as described below.
On MacOS: ~/Library/Application\ Support/Claude/claude_desktop_config.json
On Windows: %APPDATA%/Claude/claude_desktop_config.json
Note: You might need to use the direct path to uv
. Use which uv
to find the path.
Development/Unpublished Servers Configuration
"mcpServers": {
"mcp-pinecone": {
"command": "uv",
"args": [
"--directory",
"{project_dir}",
"run",
"mcp-pinecone"
]
}
}
Published Servers Configuration
"mcpServers": {
"mcp-pinecone": {
"command": "uvx",
"args": [
"--index-name",
"{your-index-name}",
"--api-key",
"{your-secret-api-key}",
"mcp-pinecone"
]
}
}
You can sign up for a Pinecone account here.
Create a new index in Pinecone, replacing {your-index-name}
and get an API key from the Pinecone dashboard, replacing {your-secret-api-key}
in the config.
To prepare the package for distribution:
- Sync dependencies and update lockfile:
uv sync
- Build package distributions:
uv build
This will create source and wheel distributions in the dist/
directory.
- Publish to PyPI:
uv publish
Note: You'll need to set PyPI credentials via environment variables or command flags:
- Token:
--token
orUV_PUBLISH_TOKEN
- Or username/password:
--username
/UV_PUBLISH_USERNAME
and--password
/UV_PUBLISH_PASSWORD
Since MCP servers run over stdio, debugging can be challenging. For the best debugging experience, we strongly recommend using the MCP Inspector.
You can launch the MCP Inspector via npm
with this command:
npx @modelcontextprotocol/inspector uv --directory {project_dir} run mcp-pinecone
Upon launching, the Inspector will display a URL that you can access in your browser to begin debugging.
This project is licensed under the MIT License. See the LICENSE file for details.
The source code is available on GitHub.
Send your ideas and feedback to me on Bluesky or by opening an issue.