DevHub
Self-hosted operational dashboard and MCP server that catalogs runnable services and their operational context, offering a read-only MCP endpoint to list projects, service status, runbooks, and reconciliation context.
README
DevHub
Git remembers the code. DevHub remembers how it runs.
Your coding agent can build and deploy it. DevHub helps you understand it, trust it, operate it, afford it and recover it.

DevHub is the self-hosted operational home for everything you build: local dashboards, APIs, workers, bots, databases, model runtimes and private tools spread across laptops, servers and private networks.
It keeps every service findable, explains how trustworthy its current state is, and leads to the safest useful next action. The catalog does not depend on a particular framework, process manager or hosting platform.
Your coding agent may have shipped the service, but DevHub preserves the operating context: where it was deployed, what it depends on, who owns it, what it may cost, which safeguards have evidence and whether recovery has actually been reviewed.
It keeps reviewed YAML manifests in Git, builds a lightweight read-only dashboard and exposes the same catalog through a read-only MCP endpoint. It does not scan random ports, execute shell commands or act as a process supervisor.
Alpha: the public snapshot is intended for evaluation and small self-hosted installations. The manifest contract may still gain backward-compatible fields before 1.0.
What it answers
- Which projects and runnable services exist?
- On which host does each service belong?
- Is an endpoint live, reported, catalog-only or on another computer?
- Which reviewed runbook explains how to start, inspect or recover it?
- Which monitoring, recovery, security, ownership and cost claims have fresh evidence, and which remain honestly unknown?
- Which services across the portfolio need an evidence refresh or a recovery decision first?
The core promise is continuity: return to a project after weeks or months and
recover its operational context in under 30 seconds. DevHub is not a metrics
stack or a universal process supervisor; when trustworthy evidence is missing,
it reports unknown instead of inventing a green status.
Refresh reviewed provider evidence
DevHub v0.7 includes narrow read-only GitHub adapters. Start from the fictional
config/evidence-bindings/example-release.json, replace its identities with a
repository and service already reviewed in your catalog, then run:
npm run devhub -- collect-evidence config/evidence-bindings/example-release.json --json
npm run devhub -- review-portfolio --json \
--evidence-binding config/evidence-bindings/example-release.json
The result is a candidate, not an automatic catalog update. Review a minimal YAML diff before dashboard or MCP can show the new evidence.
Quick start with Docker
Requirements: Docker Engine with Compose v2.
docker compose -f deploy/docker/compose.yaml up --build -d
Open http://127.0.0.1:3000. The image is built from the included demo catalog and runs as a non-root user with a read-only filesystem. Edit the catalog before using DevHub for real infrastructure, then rebuild the image. Container health verifies both the dashboard and read-only MCP initialization.
The application has no built-in user authentication. Keep the default loopback binding, or put it behind a private network or an authenticated reverse proxy. Do not expose it directly to the public internet.
Run from source
Requirements: Node.js 22.13 or newer and npm.
npm ci
npm run devhub -- validate --check
npm run dev
For a production-style check:
npm run build
npm test
npm run lint
Add your catalog
Start a separate catalog without editing or deleting the included demo:
npm run devhub -- init-catalog ./my-catalog \
--host-id developer-laptop \
--host-name "Developer laptop" \
--host-kind mac \
--host-location local
The default is a read-only plan showing every path. Review it, repeat with
--apply, then use DEVHUB_CATALOG_DIR="$PWD/my-catalog" for catalog commands
and the DevHub process. The initializer refuses non-empty destinations and
validates the created hosts.yaml plus empty projects/ starter immediately.
- Add one project file per project under
my-catalog/projects/. - Never put passwords, tokens, private keys, cookies or secret-bearing URLs in a manifest.
- Run
DEVHUB_CATALOG_DIR="$PWD/my-catalog" npm run devhub -- validateto regenerate the reviewed runtime JSON. - Commit the YAML and generated JSON together.
Use registration: native when a repository you control should own
.devhub/project.yaml. Use registration: overlay when the metadata is
private or the source repository is shared or external.
See Installation, Configuration, App Passport, Portfolio review, Read-only evidence adapters, GitHub evidence adapters, and Privacy.
Codex users can install the portable workflow skill and connect it to their own MCP endpoint using the Codex integration guide.
Read-only MCP
The same process serves Streamable HTTP MCP at /mcp. It can list and search
projects, return service status and runbooks, and prepare reconciliation
context. It cannot mutate manifests, probe caller-supplied URLs or execute
commands.
Only connect clients you trust: MCP exposes the same operational metadata as the dashboard and uses the deployment's network/authentication boundary.
Project status
DevHub is pre-1.0 software. Good evaluation feedback includes installation failures, confusing status semantics, portability gaps and catalog examples that cannot be represented safely.
License
Apache License 2.0. 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。