NTLx/ServerFS_MCP / skills/serverfs-tunnel
serverfs-tunnel
VerifiedOperate optional OpenAI Secure MCP Tunnel and Linux/macOS/Windows Cloudflare ingresses while preserving independent service owners.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 9.4 KB |
| references/cloudflare-linux.md | markdown | 3.4 KB |
| references/cloudflare-windows.md | markdown | 4.9 KB |
| references/linux.md | markdown | 1.7 KB |
| references/macos.md | markdown | 2.6 KB |
| references/windows.md | markdown | 2.0 KB |
references/cloudflare-linux.md
View on GitHub# Cloudflare Tunnel on Linux / Docker Compose (v0.24.0 stable)
This is **not** the OpenAI Secure MCP Tunnel. Do not use OpenAI control-plane credentials, client bootstrap, or health commands here.
## Read before action
- `docs/cloudflare-tunnel.md` — canonical operator instructions and live-acceptance evidence requirements.
- `compose.cloudflare.yml`, `compose.yml` and `.env.example` — actual deployment semantics.
- `docs/api-key.md` / `docs/oauth.md` — caller admission and its limitations.
- Cloudflare documentation for the current token-file, routing and networking behavior.
## Steps
1. Verify this is a Linux Compose installation. Locate the deployment and enumerate its **active** Compose file set without revealing secrets. Confirm the requested Cloudflare Tunnel is separate from any live OpenAI Tunnel or host-managed cloudflared.
2. Confirm the Cloudflare account/Zone/Tunnel and intended hostname by read-only account API or operator evidence before provisioning. Changes to DNS/public routes require explicit scope and must never overwrite an unrelated hostname.
3. Keep the remotely-managed Tunnel Token outside the checkout and all ServerFS workdirs. Store only its absolute path under `CLOUDFLARE_TUNNEL_TOKEN_FILE`; the file must be mode 0600 and readable by configured `CLOUDFLARED_UID` (independent of `SERVERFS_UID`). Never emit the token.
4. Bind the connector only to `serverfs_internal` and `tunnel_egress`. The MCP service also uses its existing `serverfs_file_ingress` network, which remains `internal:true`; do not attach MCP to `tunnel_egress`. Use Cloudflare service `http://serverfs-mcp:8000` (or configured container port), **not** 127.0.0.1, when both services run as containers.
5. Require exact ServerFS public Host allowlist and independent ServerFS API Key/OAuth. Prefer OAuth/Bearer; query-key compatibility has URL-leakage risks. Do not disable DNS rebinding or log full URLs.
6. For an existing Agent/Repo Ops deployment, preserve the same Compose overlay chain and service mounts; only create/recreate the connector and necessary MCP container when authorized.
7. Validate each layer: safe Compose render, Cloudflare-connected connector, DNS/HTTPS, valid MCP initialize (JSON or SSE response), `notifications/initialized` before persistent GET SSE, session/tools, read-only/hidden-deny policy, long-running streaming and same-session recovery after an isolated Connector restart. Cloudflare connected status alone is insufficient. Record unverified portions explicitly.
8. Record the exact scope of client E2E evidence: the VPS isolated read-only staging connection passed direct ChatGPT tool calls, SSE hold and same-session reconnect before v0.24.0 publication. Do **not** generalize this result to production workdirs, Agent/Repo Ops, or native macOS/Windows acceptance; native macOS shipped in v0.24.1. v0.24.2 adds Windows Cloudflare; its separate-owner public ChatGPT tools and cross-ingress file-operation acceptance is documented in the Windows runbook.
## Stop conditions
- No credential-safe external file path or incorrect file ownership/mode.
- No independent public caller authentication.
- Ambiguous Tunnel ownership, hostname conflict, unauthorized Cloudflare API write or unverified account.
- Deployment would expose an MCP port publicly, remove an Agent/Repo Ops overlay, or relax Host/Origin policy.
- Only an OpenAI Tunnel install is available and no Cloudflare mode was requested.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |