MCP Setup

Connect AI assistants (Claude Code, Cursor, Claude Desktop, etc.) to Revefi using the Model Context Protocol (MCP).

Overview

Revefi exposes a Model Context Protocol (MCP) HTTP endpoint that lets MCP-compatible clients (Claude Code, Cursor, Claude Desktop, etc.) call Revefi tools — query metadata, operate on warehouse data — directly from the assistant.

It authenticates using the same API access token you use for the Public APIs, and operates against the warehouse configured for your tenant (Snowflake or Databricks).

Prerequisites

  • An API access token. Follow the Public APIs setup to generate one. The token must be READ_WRITE.
  • Currently only supports Snowflake and Databricks.

Endpoint

EnvironmentURL
Productionhttps://gateway.revefi.com/api/v1/mcp
AUhttps://au.gateway.revefi.com/api/v1/mcp

The MCP server speaks the Streamable HTTP transport. Authentication is Authorization: Bearer <token> on every request.

Client setup

Below are example configurations for common MCP clients. In all cases, paste the access token from the Public APIs setup as the Authorization header value.

Claude Code

Add the server to your user-level claude config under mcpServers:

{
  "mcpServers": {
    "revefi": {
      "type": "http",
      "url": "https://gateway.revefi.com/api/v1/mcp",
      "headers": {
        "Authorization": "Bearer <PASTE_YOUR_API_TOKEN_HERE>"
      }
    }
  }
}

Restart Claude Code. The Revefi tools should appear under the revefi MCP server.

Security

Treat the access token like a password. It grants the same permissions as the Public APIs (read/write within the tenant).