DevPanel MCP Server
A local MCP server for managing SaaS projects with planning, operations, and secrets, designed to work with Claude Code via stdio. It provides tools for kanban, health monitoring, and secure secrets management, all logged for AI transparency.
README
DevPanel
A local project panel for SaaS builds — planning and ops in one place. Single
user, localhost-first, dark sci-fi/HUD UI. See SPEC.md for the full design.
A left sidebar switches between five views:
- Plan — kanban (Todo / Pågår / Blockerad / Klar), task cards with prio / AI / dependency badges, milestones, decisions, idea inbox. 🕐 on a card creates a reminder.
- Ops — service health grid (uptime / latency / cost), open-incident banner + runbooks, deploys, cert status, backup age, cost-trend sparklines.
- Filer — per-project file manager: folder tree, create
.txt/.md, inline editor (save = new version), version history + restore, drag-and-drop upload, streamed download. Blobs live in SQLite. - AI — Ollama-driven suggestions (prioritize / new task / move / question). Accept executes the action; questions get a free-text answer fed back as context.
- Inställningar — Ollama model + auto-analyze toggle, SMTP form (Gmail preset + test button).
Plus a Secrets modal (AES-256-GCM at rest, key from a master password via
argon2id — AI/MCP never see values, only {name, scope, age}) and a global
reminders widget with badge + toast for due items.
Three processes: api (Hono, serves the built web UI), worker (health poll +
daily alerts + 30-min AI auto-analyze + per-minute reminders), mcp (stdio,
spawned by Claude Code). One SQLite DB. Ollama runs locally (127.0.0.1:11434).
Architecture
- api — Hono on
127.0.0.1:8899, REST mirror of the MCP tools + secrets endpoints, and servesweb/dist. - worker — polls each service's
health_urlevery 60s; daily cron alerts on cert expiry, secret age, stale backups and cost overruns (via ntfy). - mcp — the same tools over stdio for Claude Code; every AI write is logged to
activity_logtaggedsource='ai'. - web — Vite + React SPA, built to
web/distand served by the api.
Setup
# 1. config
cp .env.example .env # tweak if needed; defaults are fine for localhost
# 2. install + build
npm install
cd web && npm install && npm run build && cd ..
# 3. run (pm2)
pm2 start ecosystem.config.cjs
pm2 save # persist across reboots (after `pm2 startup`)
Open http://127.0.0.1:8899. The database is created on first run at
data/panel.db (override with DB_PATH).
Dev mode with hot reload: cd web && npm run dev (proxies /api to the running
api on 8899).
Remote access (optional)
The api binds loopback only. To reach it remotely, front it with Cloudflare
Access over a tunnel — see cloudflared/config.example.yml.
SSH port-forwarding (ssh -L 8899:127.0.0.1:8899 host) also works.
Configuration
All configuration is via .env (see .env.example) — no
hardcoded paths. Nothing secret lives in any config file; the master password is
entered at unlock time and only its derived key is held in memory, for 5 minutes.
- Ollama (
OLLAMA_URL,OLLAMA_MODEL) — local only; auto-analyze is toggled at runtime in Inställningar (default off). - Email (
SMTP_HOST/PORT/USER/PASS/FROM) — for reminders. Prefer setting the password in.envover the UI (a UI-saved password is stored in the DB in plaintext, redacted only from API responses).
Security notes
- Secret values never appear in
get_context, logs, or API responses without an unlocked session. - Bind
127.0.0.1; expose only via Cloudflare Access or an SSH tunnel. Ollama is never exposed externally. - MCP
read_fileserves text mimes only, ≤100kb, and never secrets. .env,data/, and tunnel credentials are git-ignored — keep them that way.
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。