Use an Assistant MCP server

Every Pinecone Assistant has a dedicated MCP server that gives AI agents direct access to context from the assistant’s uploaded files through the standardized Model Context Protocol (MCP). This page shows you how to connect an assistant’s MCP server with Cursor, Claude Desktop, and LangChain.

Pinecone also provides an MCP server for managing indexes, upserting data, and querying your Pinecone database directly from an AI agent. See Use the Pinecone MCP server.

There are two ways to connect to an assistant MCP server:

Remote MCP server - Use a dedicated MCP endpoint to connect directly to an assistant.
Local MCP server - Run a Docker container locally that connects to an assistant

Both options support a context tool that allows agents to retrieve relevant context snippets from your assistant’s uploaded files. This is similar to the context API but fine-tuned for MCP clients. Additional capabilities, such as file access, will be added in future releases.

Remote MCP server

Every Pinecone Assistant has a dedicated MCP endpoint that you can connect directly to your AI applications. This option doesn’t require running any infrastructure and is managed by Pinecone. The MCP endpoint for an assistant is:

https://<YOUR_PINECONE_ASSISTANT_HOST>/mcp/assistants/<YOUR_ASSISTANT_NAME>

The previous SSE-based endpoint (with /sse suffix) is deprecated and will stop working on August 31, 2025 at 11:59:59 PM UTC. Before then, update to the streamable HTTP transport MCP endpoint shown above, which implements the current MCP specification and provides improved flexibility and compatibility.

Prerequisites

Before you begin, make sure you have the following values, which you’ll use in the commands below:

<YOUR_PINECONE_API_KEY>: A Pinecone API key.
<YOUR_PINECONE_ASSISTANT_HOST>: In the Pinecone console, this is your assistant’s Host value.
<YOUR_ASSISTANT_NAME>: Your assistant’s name, as displayed in the Pinecone console. For example, example-assistant.

Use with Claude Code

You can use the Claude CLI to configure Claude Code to use your assistant’s remote MCP server. For more information, see Claude Code’s MCP documentation.

Add the MCP server using the Claude CLI:
claude mcp add --transport http my-assistant https://<YOUR_PINECONE_ASSISTANT_HOST>/mcp/assistants/<YOUR_ASSISTANT_NAME> --header "Authorization: Bearer <YOUR_PINECONE_API_KEY>"
Replace <YOUR_PINECONE_API_KEY> with your Pinecone API key, <YOUR_PINECONE_ASSISTANT_HOST> with your Pinecone Assistant host, and <YOUR_ASSISTANT_NAME> with your assistant’s name.
Verify the server was added successfully:
claude mcp get my-assistant
The MCP server tools should now be available in Claude Code’s chat interface.

Use with Claude Desktop

You can configure Claude Desktop to use your assistant’s remote MCP server. However, at this early stage of remote MCP server adoption, the Claude Desktop application does not support remote server URLs. In the example below, we work around this by using a local proxy server, supergateway, to forward requests to the remote MCP server with your API key.

supergateway is an open-source third-party tool. Use at your own risk.

Open Claude Desktop and go to Settings.
On the Developer tab, click Edit Config to open the configuration file.

Add the following configuration:

{
  "mcpServers": {
    "Assistant over supergateway": {
      "command": "npx",
      "args": [
        "-y",
        "supergateway",
        "--streamableHttp",
        "https://<YOUR_PINECONE_ASSISTANT_HOST>/mcp/assistants/<YOUR_ASSISTANT_NAME>",
        "--header",
        "Authorization: Bearer <YOUR_PINECONE_API_KEY>"
      ]
    }
  }
}

Replace <YOUR_PINECONE_API_KEY> with your Pinecone API key and <YOUR_PINECONE_ASSISTANT_HOST> with your Pinecone Assistant host.

Save the configuration file and restart Claude Desktop.
From the new chat screen, you should see a hammer (MCP) icon appear with the new MCP server available.

Use with Cursor

You can configure Cursor to use your assistant’s remote MCP server directly through the .cursor/mcp.json configuration file.

Open Cursor and create a .cursor directory in your project root if it doesn’t exist.
Open .cursor/mcp.json (create it if necessary). To learn more, refer to Cursor’s MCP documentation.

Add the following configuration:

{
  "mcpServers": {
    "pinecone-assistant": {
      "url": "https://<YOUR_PINECONE_ASSISTANT_HOST>/mcp/assistants/<YOUR_ASSISTANT_NAME>",
      "headers": {
        "Authorization": "Bearer <YOUR_PINECONE_API_KEY>"
      }
    }
  }
}

Replace <YOUR_PINECONE_API_KEY> with your Pinecone API key, <YOUR_PINECONE_ASSISTANT_HOST> with your Pinecone Assistant host, and <YOUR_ASSISTANT_NAME> with your assistant’s name.

Save the configuration file.
The MCP server tools should now be available in Cursor’s chat interface.

Use with LangChain

You can use the LangChain MCP client to integrate with LangChain to create a powerful multi-agent workflow. For example, the following code integrates Langchain with two assistants, one called ai-news and the other called industry-reports:

Python

# Example code for integrating with LangChain
from langchain_mcp_adapters.client import MultiServerMCPClient
from langgraph.prebuilt import create_react_agent

from langchain_anthropic import ChatAnthropic

model = ChatAnthropic(model_name="claude-3-7-sonnet-latest", api_key="YOUR_ANTHROPIC_API_KEY")
pinecone_api_key = "<YOUR_PINECONE_API_KEY>"

async with MultiServerMCPClient(
    {
        "assistant_ai_news": {
            "url": "https://prod-1-data.ke.pinecone.io/mcp/assistants/ai-news",
            "transport": "streamable_http",
            "headers": {
                "Authorization": f"Bearer {pinecone_api_key}"
            }
        },
        "assistant_industry_reports": {
            "url": "https://prod-1-data.ke.pinecone.io/mcp/assistants/industry-reports",
            "transport": "streamable_http",
            "headers": {
                "Authorization": f"Bearer {pinecone_api_key}"
            }
        }
    }
) as client:
    agent = create_react_agent(model, client.get_tools())

    response = await agent.ainvoke({"messages": "Your task is research the next trends in AI, and form a report with the most undervalued companies in the space. You have access to two assistants, one that can help you find the latest trends in AI, and one that can help you find reports on companies."})
    print(response["messages"][-1].content)

Local MCP server

Pinecone provides an open-source Pinecone Assistant MCP server that you can run locally with Docker. This option is useful for development, testing, or when you want to run the MCP server within your own infrastructure or expand the MCP server to include additional capabilities. For the most up-to-date information on the local MCP server, see the Pinecone Assistant MCP server repository.

Prerequisites

Docker is installed and running on your system.
A Pinecone API key. You can create a new key in the Pinecone console.
Your Pinecone Assistant host. To find it, go to your assistant in the Pinecone console. You’ll see the assistant Host in the sidebar.

Start the MCP server

Download the assistant-mcp Docker image:

docker pull ghcr.io/pinecone-io/assistant-mcp

Start the MCP server, providing your Pinecone API key and Pinecone Assistant host:

docker run -i --rm \
  -e PINECONE_API_KEY=<PINECONE_API_KEY> \
  -e PINECONE_ASSISTANT_HOST=<PINECONE_ASSISTANT_HOST> \
  pinecone/assistant-mcp

Use with Claude Desktop

Open Claude Desktop and go to Settings.
On the Developer tab, click Edit Config to open the configuration file.

Add the following configuration:

{
  "mcpServers": {
    "pinecone-assistant": {
      "command": "docker",
      "args": [
        "run", 
        "-i", 
        "--rm", 
        "-e", 
        "PINECONE_API_KEY", 
        "-e", 
        "PINECONE_ASSISTANT_HOST", 
        "pinecone/assistant-mcp"
      ],
      "env": {
        "PINECONE_API_KEY": "<YOUR_PINECONE_API_KEY>",
        "PINECONE_ASSISTANT_HOST": "<YOUR_PINECONE_ASSISTANT_HOST>"
      }
    }
  }
}

Replace <YOUR_PINECONE_API_KEY> with your Pinecone API key and <YOUR_PINECONE_ASSISTANT_HOST> with your Pinecone Assistant host.

Save the configuration file and restart Claude Desktop.
From the new chat screen, you should see a hammer (MCP) icon appear with the new MCP server available.

Use with Cursor

Open Cursor and create a .cursor directory in your project root if it doesn’t exist.
Open .cursor/mcp.json (create it if necessary). To learn more, refer to Cursor’s MCP documentation.

Add the following configuration:

{
  "mcpServers": {
    "pinecone-assistant": {
      "command": "docker",
      "args": [
        "run", 
        "-i", 
        "--rm", 
        "-e", 
        "PINECONE_API_KEY", 
        "-e", 
        "PINECONE_ASSISTANT_HOST", 
        "pinecone/assistant-mcp"
      ],
      "env": {
        "PINECONE_API_KEY": "<YOUR_PINECONE_API_KEY>",
        "PINECONE_ASSISTANT_HOST": "<YOUR_PINECONE_ASSISTANT_HOST>"
      }
    }
  }
}

Replace <YOUR_PINECONE_API_KEY> with your Pinecone API key and <YOUR_PINECONE_ASSISTANT_HOST> with your Pinecone Assistant host.

Save the configuration file.

Next Steps

Visit the Pinecone Assistant MCP Server repository for detailed installation and usage instructions
Learn about Model Context Protocol and how it enables AI agents to interact with tools and data
Explore retrieve context snippets to understand the underlying API functionality

Get started

Build an assistant

Upload your data

Chat with an assistant

Evaluate answers

Retrieve context snippets

Integrate with AI agents

Admin

Use an Assistant MCP server

Remote MCP server

Prerequisites

Use with Claude Code

Use with Claude Desktop

Use with Cursor

Use with LangChain

Local MCP server

Prerequisites

Start the MCP server

Use with Claude Desktop

Use with Cursor

Next Steps

​Remote MCP server

​Prerequisites

​Use with Claude Code

​Use with Claude Desktop

​Use with Cursor

​Use with LangChain

​Local MCP server

​Prerequisites

​Start the MCP server

​Use with Claude Desktop

​Use with Cursor

​Next Steps

Remote MCP server

Prerequisites

Use with Claude Code

Use with Claude Desktop

Use with Cursor

Use with LangChain

Local MCP server

Prerequisites

Start the MCP server

Use with Claude Desktop

Use with Cursor

Next Steps