Cortex_MCP_Server

sarathi-aiml/Cortex_MCP_Server

3.1

If you are the rightful owner of Cortex_MCP_Server and would like to certify it and/or have it hosted online, please leave a comment on the right or send an email to henry@mcphub.com.

The Model Context Protocol (MCP) server acts as a bridge between AI assistants and external tools, enabling seamless integration and communication.

MCP Server for Cortex Agents Setup

Run a local MCP server that exposes Snowflake Cortex Agents tools, then wire it into Claude Desktop.

What is MCP?

MCP (Model Context Protocol) is a lightweight protocol that lets external tools and services connect to AI assistants like Claude.
It works as a bridge between the AI and your own systems.

MCP vs Agents

  • MCP: Exposes tools. It does not decide how to use them.
  • Agents: Run reasoning loops, choose tools, and handle workflows. MCP gives agents the tools they need.

1) Prerequisites

  • macOS with Terminal and zsh.
  • Snowflake account with access to Cortex Search and Cortex Analyst.
  • Programmatic Access Token (PAT) in Snowflake.
  • Claude Desktop installed.

Follow the Quickstart for Cortex Agents and create a Cortex Search service and a Cortex Analyst in your Snowflake account:
https://quickstarts.snowflake.com/guide/getting_started_with_cortex_agents/

2) Install uv

uv runs Python projects fast without manual virtualenv setup.

Pick one method.

Option A, Homebrew

brew install uv

Option B, official script

curl -LsSf https://astral.sh/uv/install.sh | sh

Add uv to PATH

If uv --version fails, add the install path.

  • For Homebrew on Apple Silicon:
echo 'export PATH="/opt/homebrew/bin:$PATH"' >> ~/.zshrc
source ~/.zshrc
  • For installer script (default path):
echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.zshrc
source ~/.zshrc

Verify

uv --version

You should see a version number.

3) Prepare the project folder

  • Download the code from the link below and place it in the project folder:
    https://github.com/sarathi-aiml/Cortex_MCP_Server

  • Place your MCP server code in a folder. Example:
    /Users/<you>/Downloads/MCP/mcp-server-cortex

  • Create PAT in Snowflake and add it to the .env.dev file.

  • Update .env.dev in your local machine with the following values:

SNOWFLAKE_ACCOUNT_URL=https://<account>.snowflakecomputing.com
SNOWFLAKE_PAT=<your_pat_here>
CORTEX_SEARCH_SERVICE=<your_search_service_name>
SEMANTIC_MODEL_FILE=<path_or_identifier>

4) Install Python requirements

Create a file named requirements.txt in your project folder with:

httpx
python-dotenv
mcp

Install them with:

uv pip install -r requirements.txt

5) Run the MCP server

Run the server using:

uv run mcp_cortex_server.py

or

python mcp_cortex_server.py

You will not see output in the terminal, but the server is running.

6) Wire it into Claude Desktop

  1. Open Claude Desktop.
  2. Go to Settings → Developer.
  3. Click Edit config or Add config. Finder opens claude_desktop_config.json. Open this file in any code editor.
  4. Insert a block like this. Adjust paths to your user folder.
{
  "mcpServers": {
    "cortex-agent": {
      "command": "/Users/<you>/.local/bin/uv",
      "args": [
        "--directory",
        "/Users/<you>/Downloads/MCP/mcp-server-cortex",
        "run",
        "mcp_cortex_server.py"
      ]
    }
  }
}

Make sure "command" points to the correct uv path and "args" has the right path to mcp_cortex_server.py.

  1. Save the file.
  2. Quit Claude Desktop completely (not just close).
  3. Reopen Claude Desktop.
  4. In a chat, click the Tools icon next to the + button. You should see cortex-agent.

7) Test the server from Claude Desktop

Sample prompts:

  • "What is the total deal value for each product line? Also find top performing product and list the reason. Use Cortex agents."
  • "What is the total deal value for each product line? Use Cortex agents."

8) MCP Server Flow

Claude Desktop ---> MCP Config ---> Cortex MCP Server ---> Snowflake API
                                           |  
                                           +--> Cortex Analyst (SQL generation)
                                           +--> Cortex Search (semantic search)
                                           +--> SQL Exec (runs in Snowflake)

9) Troubleshooting

  • uv not found → Fix PATH as in section 2, restart terminal, run uv --version.
  • No cortex-agent in Tools → Recheck claude_desktop_config.json syntax, restart Claude Desktop.
  • 401/403 from Snowflake → Verify PAT is valid, check role/warehouse/network policy.
  • Cortex Search errors → Confirm the service name matches .env.dev, check index status.
  • Cortex Analyst errors → Verify semantic model file path or identifier.
  • SSE stalls → Try different network if proxies block event streams.

10) Security tips

  • Store secrets in .env.dev with restricted permissions.
  • Rotate PAT regularly.
  • Do not print secrets in logs.

11) Summary checklist

  • uv installed and in PATH.
  • .env.dev contains SNOWFLAKE_ACCOUNT_URL, SNOWFLAKE_PAT, CORTEX_SEARCH_SERVICE, SEMANTIC_MODEL_FILE.
  • requirements.txt installed.
  • Snowflake Cortex Search and Analyst are available.
  • Claude Desktop config points to uv and the project folder.
  • Restarted Claude. cortex-agent visible and working.

Best Practices for MCP Server Setup

  • Use TLS and authentication to secure communication.
  • Keep services modular, stateless, and independently deployable.
  • Implement monitoring, structured logging, and trace IDs for visibility.
  • Design for scalability and graceful degradation under load.
  • Add health checks and auto-restart policies for resilience.
  • Regularly rotate PAT tokens and review Snowflake roles.
  • Test with mock queries and sandbox data before production.
  • Apply resource limits (CPU, memory) to prevent overload.
  • Document tool contracts (inputs, outputs, failure modes) for easier extension.
  • Run MCP servers in isolated environments (containers, VMs, or sandboxes).