Multi-Hoster-Upload/docs/remote-diagnostics-setup.md
Administrator d69e5c39bf feat(diagnostics): MCP gateway + harden redaction so no secret ever leaves the box
Adds the connect-by-code side of remote diagnostics and closes two real
secret-leak vectors that an end-to-end gateway<->agent test surfaced.

Gateway (gateway/, local stdio MCP, Claude connects once):
- 14 read-only tools (server_health hub, read_log, list_logs, list_errors,
  get_queue_state, get_history, get_config_redacted, get_system_info,
  get_rotation_state, get_app_events + connect/disconnect/list/current).
- The HOST is always supplied by the operator, never taken from the code.
- TLS fingerprint pinning is enforced in the socket 'open' handler BEFORE the
  token is sent (wss opt-in); plain ws is loopback-only.
- registry.json (holds bearer tokens) is gitignored; only an empty example ships.

Security hardening (gates every off-box payload):
- redactLogText now scrubs opaque bearer/token-family secrets that are NOT
  stored config credentials (e.g. a session token a hoster returns inside an
  error string): bare token/auth_token/refresh_token/session_token + standalone
  "Bearer <opaque>". Benign "token bucket" prose is left intact.
- get_config_redacted deep-redacts every string leaf (JSON-safe, per-leaf, so
  the cookie/sess line patterns can't gobble across a compact-JSON field) and
  drops the history subtree (served by get_history with its own per-error
  redaction). This plugs leaks via globalSettings.pendingQueue[].error etc.

Bind-address safety:
- _safeDiagBindAddress() forces the diagnostic agent to 127.0.0.1/::1; the
  0.0.0.0 UI option is removed. Direct LAN/Internet bind stays disabled until
  encrypted transport (wss) exists — remote access goes through an SSH/VPN
  tunnel to loopback. (Never plaintext ws:// on all interfaces.)

Tests: end-to-end gateway<->agent gate (connect -> server_health/read_log/
get_config_redacted, asserts zero secret leakage, rejects doodstream log, path
traversal and write ops); + redaction regression tests in the main suite.
385 app tests + 9 gateway tests pass; lint 0 errors.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-06-19 17:39:31 +02:00

127 lines
4.9 KiB
Markdown

# Remote Diagnostics Setup (read-only)
This guide explains how to let Claude Code run **read-only** diagnostics against a
Multi-Hoster-Uploader instance running on a remote Windows server, through the local
**stdio MCP gateway** in `gateway/`.
The gateway is an MCP server to Claude and a WebSocket client to a diagnostic agent inside the
app. It never touches the screen, never injects input, and never writes anything on the server.
---
## 1. Enable diagnostics on the server and copy the code
On the **server** (the machine running the app):
1. Open the app's **Settings**.
2. Enable **"Diagnose-Zugriff"**.
3. Copy the connection **code**. It looks like `mhu1_<base64url...>`.
The code carries a one-time auth **token** (and, for TLS, the server cert fingerprint). It does
**not** carry a host — you supply the host yourself when you connect. Treat the code as a
**secret**: anyone with the code and network reach to the agent can read diagnostics.
The agent's **safe default is to bind to `127.0.0.1`** (loopback only). It is not exposed to the
network. You reach it through a tunnel (next step).
---
## 2. Reach the agent over a tunnel (SSH local port-forward or WireGuard)
Because the agent binds to `127.0.0.1` on the server, open a tunnel from your machine to the
server's loopback. The agent's default port is `9110`.
### SSH local port-forward (recommended)
```
ssh -L 9110:127.0.0.1:9110 user@server
```
Leave that session open. Now `127.0.0.1:9110` on **your** machine is forwarded to
`127.0.0.1:9110` on the **server**. You connect Claude to **`127.0.0.1`** (your local end of the
tunnel), not the server's public IP.
### WireGuard (alternative)
Bring up a WireGuard tunnel to the server, then connect to the server's WireGuard address (or, if
you forward loopback over the tunnel, `127.0.0.1`). Use whichever address resolves to the agent's
`127.0.0.1:9110` on the server.
> Only expose the agent directly on a LAN/VPN IP if you fully trust that network segment. The
> default loopback + tunnel is the secure choice.
---
## 3. Install the gateway and register it with Claude (one time)
On the machine running Claude Code:
```
cd gateway
npm install
```
Register the gateway as an MCP server (one time):
```
claude mcp add --transport stdio mhu-diag -- node "C:\Users\ploet\Desktop\Claude Projekte\multi-hoster-uploader\gateway\index.js"
```
Adjust the absolute path if your checkout lives elsewhere.
---
## 4. Usage
With the tunnel up, tell Claude:
```
server prod-3 at 127.0.0.1, code mhu1_<...>
```
Claude will:
1. call `connect_server(code:"mhu1_...", host:"127.0.0.1")`,
2. immediately read the app version via `get_system_info`,
3. remember the server under its label in `registry.json`.
Next time you can just say:
```
connect_server(label:"prod-3")
```
with no code — the gateway dials it from the registry. Then ask Claude "what's wrong?" — it will
typically start with **`server_health`**, the one-shot hub that aggregates errors, queue, rotation
and system info in a single call. Other read-only tools: `read_log`, `list_logs`, `list_errors`,
`get_queue_state`, `get_history`, `get_config_redacted`, `get_rotation_state`, `get_app_events`.
---
## 5. Failure → meaning
When a connect or a request fails, the gateway returns a human-readable cause. Quick reference:
| Symptom / signal | What it means / what to do |
| -------------------------------------- | ----------------------------------------------------------------------- |
| `ECONNREFUSED` | app not running / wrong port / inbound firewall closed |
| `ETIMEDOUT` | firewall DROP / NAT not forwarded / tunnel down |
| close code **4001** | auth timeout, retry |
| close code **4002** | stale or rotated code — re-copy the current code from the server |
| close code **4003** | brute-force lockout, wait 60s |
| connected but no `auth-ok` | old app version without the diagnostic agent — update that server |
| wss fingerprint mismatch | server cert changed (reinstalled?) — re-copy the code |
If you tunnel and still get `ECONNREFUSED`, check that the SSH session is up and that the agent is
actually listening on `127.0.0.1:9110` on the server (Diagnose-Zugriff enabled).
---
## 6. What this is — and is not
- It is **READ-ONLY**. It reads logs, errors, queue/history, redacted config, rotation and system
info. **No screen capture. No input injection. No writes or config changes.**
- The **code is a secret** (it carries the auth token). Don't paste it into public channels.
- `gateway/registry.json` stores tokens for remembered servers and is **git-ignored** — never
commit it.