Stackbilt
We built a hosted MCP server that gives AI agents two capabilities through a single connection: 1. img-forge — Image generation across 5 quality tiers (SDXL Lightning to Gemini 3.1 Flash). One tool call, 5-30 seconds, done. 2. Stackbilder — Full-stack architecture scaffolding. Describe a product, get a 6-mode pipeline (Product → UX → Risk → Architect → TDD → Sprint)
README
Stackbilt MCP Gateway
<p align="center"> <img src="docs/banner.png" alt="Stackbilt MCP Gateway — img-forge + Stackbilder" width="480" /> </p>
<p align="center"> <em>Two products. One MCP connection. Image generated by img-forge (ultra tier).</em> </p>
OAuth-authenticated Model Context Protocol (MCP) gateway for Stackbilt platform services. Built as a Cloudflare Worker using @cloudflare/workers-oauth-provider.
What It Does
A single MCP endpoint (mcp.stackbilt.dev/mcp) that routes tool calls to multiple backend product workers:
| Backend | Tools | Description |
|---|---|---|
| Stackbilder | flow_create, flow_status, flow_summary, flow_quality, flow_governance, flow_advance, flow_recover |
Architecture flow orchestration |
| img-forge | image_generate, image_list_models, image_check_job |
AI image generation |
Key Features
- OAuth 2.1 with PKCE — GitHub SSO, Google SSO, and email/password authentication
- Backend adapter pattern — tool catalogs aggregated from multiple service bindings, namespaced to avoid collisions
- Security Constitution compliance — every tool declares a risk level (
READ_ONLY,LOCAL_MUTATION,EXTERNAL_MUTATION); structured audit logging with secret redaction; HMAC-signed identity tokens - Coming-soon gate —
PUBLIC_SIGNUPS_ENABLEDflag to control public access - MCP JSON-RPC over HTTP — supports both streaming (SSE) and request/response transport
Quick Start
Prerequisites
- Node.js 18+
- Wrangler CLI (
npm i -g wrangler) - Cloudflare account with the required service bindings configured
Install & Run
npm install
npm run dev
Run Tests
npm test
Deploy
npm run deploy
Deploys to the mcp.stackbilt.dev custom domain via Cloudflare Workers.
Environment Variables & Secrets
| Name | Type | Description |
|---|---|---|
SERVICE_BINDING_SECRET |
Secret | HMAC-SHA256 key for signing identity tokens |
API_BASE_URL |
Variable | Base URL for OAuth redirects (e.g. https://mcp.stackbilt.dev) |
AUTH_SERVICE |
Service Binding | RPC to stackbilt-auth worker (AuthEntrypoint) |
STACKBILDER |
Service Binding | Route to edge-stack-architect-v2 worker |
IMG_FORGE |
Service Binding | Route to img-forge-mcp worker |
OAUTH_KV |
KV Namespace | Stores social OAuth state (5-min TTL entries) |
PLATFORM_EVENTS_QUEUE |
Queue | BizOps audit event pipeline (stackbilt-user-events) |
Set secrets with:
wrangler secret put SERVICE_BINDING_SECRET
Project Structure
src/
index.ts # Entry point — OAuthProvider setup, CORS, health check bypass
gateway.ts # MCP JSON-RPC transport, session management, tool dispatch
oauth-handler.ts # OAuth 2.1 flows: login, signup, social SSO, consent
tool-registry.ts # Tool catalog aggregation, namespacing, schema validation
audit.ts # Structured audit logging, secret redaction, trace IDs
auth.ts # Bearer token extraction & validation
route-table.ts # Static routing table, tool-to-backend mapping, risk levels
types.ts # Type definitions, RiskLevel enum, interfaces
test/
audit.test.ts
auth.test.ts
gateway.test.ts
oauth-handler.test.ts
route-table.test.ts
tool-registry.test.ts
docs/
user-guide.md # End-user guide: account creation, client setup, tool usage
api-reference.md # MCP tool surface, authentication flow, tool routing
architecture.md # System design, security model, request flow
Test Suite
122 tests across 6 test files covering:
- OAuth handler — identity token signing/verification, login, signup, social OAuth flows, consent, HTML escaping
- Gateway — session lifecycle,
initialize,tools/list,tools/call, SSE streaming, error handling - Audit — secret redaction patterns (API keys, bearer tokens, hex hashes, password fields), trace IDs, queue emission
- Auth — bearer token extraction, API key vs JWT validation, error mapping
- Tool registry — catalog building, name mapping, schema validation, risk level enforcement
- Route table — route resolution, risk level lookup
npm test # single run
npm run test:watch # watch mode
Documentation
- User Guide — account creation, client setup, tool usage
- API Reference — MCP tools, authentication, tool routing
- Architecture — system design, security model, data flow
License
MIT — see LICENSE
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。