MCP tunnels - Connect Claude to private MCP servers (BETA)
This Claude documentation page describes MCP tunnels, a beta feature that securely connects Claude to Model Context Protocol (MCP) servers running inside private networks without opening inbound ports or exposing services to the public internet. The architecture uses two components deployed inside your network: cloudflared (a tunnel agent that initiates outbound-only connections to Anthropic's tunnel edge) and a proxy (terminates inner TLS, validates IP ranges, and routes requests to upstream MCP servers). Traffic flows over outbound-only connections, eliminating the need for inbound firewall rules, IP allowlisting, or public exposure. Security layers include outer mTLS with IP validation, inner TLS terminating at your proxy (so the transport provider cannot read payloads), and optional OAuth on each MCP server. Prerequisites include a deployment target (Kubernetes or VM with Docker), a tunnel created in the Claude Console, authentication (programmatic via Workload Identity Federation or manual static credentials), and outbound connectivity to api.anthropic.com and Cloudflare's tunnel edge (198.41.192.0/19). The documentation includes quickstart, Helm deployment, Docker Compose deployment, and usage instructions for Managed Agents (Console) and the Messages API. The beta is provided "as-is" with no uptime or support commitment, depends on Cloudflare as a third-party network provider, and may be modified or discontinued at any time.
https://platform.claude.com/docs/en/agents-and-tools/mcp-tunnels/overview
Comments
Post a Comment