Security: Dokploy MCP responses include GitHub App secrets (private key / client secret / webhook secret)

[ddwnbot]

While using the MCP server, the application-one response contains full GitHub App secrets (private key, client secret, webhook secret). This leaks sensitive credentials into MCP responses and logs.

Example fields returned: githubPrivateKey, githubClientSecret, githubWebhookSecret.

These should never be returned to MCP clients. Please consider redacting/omitting secrets in API responses or adding a safe mode to strip them.

Impact: Any MCP client (and logs) can exfiltrate GitHub App secrets, leading to account compromise.

Request: remove/omit secret fields from responses by default, or add a server-side redaction option.

[vcode-sh]

IMO that depend. if you vibe code then ai agent should have these in response... if you know what u doing then mcp shouldn't put them in llm context... still there are much bigger problems with this mcp - token usage. that's why you should do the same like me - fork it and rewrite to code mode... https://www.npmjs.com/package/@vibetools/dokploy-mcp

[ddwnbot]

App secrets should never be put into LLM context.

Normal setups pass secrets via env vars/secret stores and execute actions through tools, not by exposing raw credentials to the model. Even in “vibe code” modes, leaking private keys is a security footgun and violates least‑privilege.

If MCP needs credentials, it should keep them server‑side and return only minimal, non‑sensitive outputs. Please add a safe mode (default) that redacts secrets and avoids injecting them into responses/context. I’d strongly recommend fixing this in upstream rather than telling users to fork.

[vcode-sh]

on my code mode dokploy mcp I filtered that out, however if user precisely ask for these secrets mcp will override procedures, find them and return.