メインコンテンツにスキップ

Connecting AI assistants safely to Coolify deployments, n8n workflows, and GitHub CI without exposing master infrastructure tokens

Infrastructure Observability with Ops Hub MCP and Out-of-Band Telegram Approvals

When a deployment stalls or an automation workflow fails, engineers shouldn't have to jump across five dashboards. Here is how we engineered Ops Hub MCP with deep-frozen policy pipelines, automatic secret redaction, and Telegram approval gates.

2026年10月28日• 8 最小読み取り• の言葉 Sajjad Samaei
Sajjad Samaei, product designer and full-stack engineer

Six-stage Ops Hub policy pipeline with allowlist target resolution and 20 KB redacted envelopes

Six-stage Ops Hub policy pipeline with allowlist target resolution and 20 KB redacted envelopes
目次

Debugging a production incident across multiple apps usually means jumping between Coolify deployment logs, n8n workflow executions, and GitHub Actions runs. Handing raw master API tokens for those services to an AI assistant is a major security risk. We built Ops Hub MCP (`packages/ops-hub`) and Ops Proxy (`packages/ops-proxy`) to give AI assistants a read-only, secret-redacted view of our entire infrastructure.

The Six-Stage Policy Pipeline

Every tool call in ops-hub passes through createPolicy: strict Zod validation, deepFreeze argument cloning, friendly-name allowlist resolution (never raw UUIDs), sliding-window rate limits, out-of-band Telegram approval for action tools (with a 5-minute lockout after two denials), and automatic secret redaction capped at 20 KB.

Learn how we build secure infrastructure and AI tooling on our services page.

記事をシェアする

関連記事

製品の製作と実行について読み続けてください。

何かプロジェクトを考えていますか?

私たちに手紙を書いてください。 24 時間以内に返信し、次のステップを一緒に決定します。