Skip to main content
MCP (Model Context Protocol) servers expose data and capabilities to LLMs through a standardized interface. This guide helps you deploy MCP servers on TrueFoundry as services that can be accessed by the AI Gateway or other applications. MCP servers can be either HTTP/SSE-based or stdio-based. HTTP/SSE-based servers: Expose HTTP endpoints directly. Deploy without additional wrappers. Stdio-based servers: Communicate via stdin/stdout. Wrap with mcp-proxy to convert to HTTP.

​
Choose Your Deployment Path

​
Deploy from Code

Use this when:
  • You have MCP server source code (GitHub repository or local)
  • You’ve written your own MCP server
  • You need to customize the server
Supports: HTTP/SSE-based servers (deploy directly) and stdio-based servers (wrapped with mcp-proxy) Deploy MCP Server from Code →

​
Deploy from npx/uvx

Use this when:
  • You have instructions for using an MCP server in VSCode/Cursor/Claude
  • The MCP server is available as an npm or Python package
  • You don’t need to modify the server code
Supports: npm packages (via npx) and Python packages (via uvx) Deploy MCP Server from npx/uvx →

​
Prerequisites

  • A TrueFoundry workspace (Create workspace if needed)
  • Access to deploy services
  • Your MCP server code or package name
If you’re new to deploying services on TrueFoundry, we recommend first going through the Deploy Your First Service guide.

​
After Deployment

  1. Note the service endpoint URL from the deployment dashboard
  2. Test the endpoint to verify it’s responding
  3. Connect to AI Gateway using the endpoint URL (see MCP Server Getting Started)