A cloud-native web application firewall (WAF) built on NGINX, with a web interface, API, and plugin system.
GitHub stars
11,242
GitHub snapshot, 2026-11-10
Open-source license
AGPL
AGPL-3.0
Documented environments
4
Linux, Docker, Swarm, Kubernetes
BunkerWebDecision 1 of 4
Context
A web application firewall can become another specialist system to configure and operate. BunkerWeb has to expose strong protection without forcing every team to become an NGINX or security expert first.
Decision
Start from secure defaults and make configuration explicit.
Consequence
A stricter first run, and more questions to answer in public whenever a safe default blocks something a team expected to work.
A Python Model Context Protocol (MCP) server that exposes the BunkerWeb API through constrained tools and resources.
Built-in API tools
43
Public repository, 2026-08-19
Transports
3
stdio, HTTP, WebSocket
Existing authority
1
The authenticated BunkerWeb API
BunkerWeb MCPDecision 2 of 4
Context
An AI client needs a narrow, reviewable interface to infrastructure. Giving it raw configuration access would make ordinary operator questions harder to validate and easier to overreach.
Decision
Use the existing authenticated API instead of adding a second privilege model.
Consequence
The server can never offer more than that API already allows, including where its limits are inconvenient.
A four-host private platform run as code, with 69 in-scope Compose stacks, bounded automation, observability, and tested configuration invariants.
Services under management
91
Defined across 69 Compose stacks; 22 project stacks excluded
Images digest-pinned
83/83
Version and manifest digest, no :latest
Config invariants tested
28
Two repo-wide analyzers, wired as pre-commit hooks
Private repository snapshot verified on 19 August 2026. No public source is available.
Private platform engineeringDecision 2 of 4
Context
The costly failures were silent: dynamically allocated addresses reclaiming pinned slots after restarts, bind-mounted files staying on old inodes, and sign-in routes selecting the wrong provider. The configuration could still render cleanly while the running system was already wrong.
Decision
Turn network allocation and sign-in routing incidents into two tree-wide analyzers backed by 28 unit tests.
Consequence
The analyzers only catch the failure classes already understood; a new class stays invisible until someone writes its analyzer.
One local OpenTelemetry stack for comparing usage, latency, failures, tool activity, traces, and estimated cost across three AI coding command-line interfaces (CLIs).
AI CLI integrations
3
Codex, Claude Code, Gemini CLI
AppIndicator HiderDecision 4 of 4
Context
Tray support is often all or nothing. Some background applications still need AppIndicator integration, but that should not force every registered icon into the top bar.
Decision
Ship with an empty rule set, strict schema compilation, and a reproducible extension package check.
Consequence
A fresh install does nothing at all, so the extension only shows its value once someone configures it.