mcp-tunnels-cloudflared
Use to let a hosted agent reach a private-data MCP server behind your firewall (an outbound tunnel plus a proxy, with per-server OAuth) so internal tools are usable without exposing them to the public internet.
- Score
- 99.3862 signals
- Evidence
- 15,474 stars · 1,417 forks
- Last commit
- as last read from GitHub; most reads are from 2 Sep 2026 or later
- Listed
Install
No one-command install. Set it up from its source.
What it is
A networking pattern for connecting hosted agents to private MCP servers. An outbound tunnel (no inbound firewall holes) plus a provider proxy lets a managed agent call MCP servers that live inside your network, with per-server OAuth for access control. It bridges "managed agent" and "private internal tools."
When to use it
When an agent runs in a hosted control plane but needs tools or data that must stay private: internal databases, company knowledge, on-prem services exposed as MCP. The trigger is "the hosted agent needs our private MCP server."
How to install / invoke
Run the tunnel agent to establish an outbound connection on its dedicated port, deploy it via your container/Helm setup, and set the appropriate MCP-client beta header so the agent reaches the tunneled servers with per-server OAuth.
Notes
Outbound-only tunnels avoid opening inbound firewall ports: the security win. Scope each MCP server's OAuth narrowly. Pair with a hosted-control runtime when the data plane stays on your infra.
Related
claude-managed-agents-selfhost
Use for enterprise hosted-control agents (the agent loop runs at the provider while execution happens on your own infrastructure) when you want a managed control plane but data and code must stay on your machines.
Not Indexed
- server-memory