steps-mcp
Task planning and execution MCP server with durable SQLite storage and a browser UI for reviewing plans and following progress.
README
Steps MCP
Steps MCP is an agent-friendly task planning and execution MCP server with durable SQLite storage and a browser UI for reviewing plans and following progress.

The board makes the execution state visible at a glance while the MCP surface gives agents compact tools, durable resources, and an explicit next safe action.
Run with Docker
docker run -d \
--name steps-mcp \
--restart unless-stopped \
-v steps-mcp-data:/app/data \
-p 3001:3001 \
ghcr.io/dassader/steps-mcp:latest
The /app/data volume stores plans, steps, notes, transitions, attachments, and workflow state.
The latest image is published for linux/amd64 and linux/arm64, so Docker selects the correct platform automatically.
After the container starts:
- MCP endpoint:
http://localhost:3001/mcp - User interface:
http://localhost:3001/ - Health check:
http://localhost:3001/health
Connect an MCP client to the Streamable HTTP endpoint above. Call server.get_started for the built-in workflow guide, then use the returned prompts, tools, and resource URIs to plan and execute work.
Workflow
Steps MCP keeps planning, review, execution, verification, and blockers explicit:
- Create a draft plan and detailed steps.
- Review the plan in the browser UI and record explicit approval.
- Start the server-selected next step.
- Add durable notes and attachments while work progresses.
- Move completed implementation through verification, or record an honest blocker.
Tool responses include the current state, relevant identifiers, and the next safe action so an agent does not need the entire state machine in tool descriptions.

Durable work evidence
Every status transition creates an audit note. Agents can also add standalone notes and attach text, binary evidence, or links without changing step status.

Configuration
| Variable | Default | Purpose |
|---|---|---|
HOST |
0.0.0.0 in Docker |
HTTP bind address |
PORT |
3001 |
Container HTTP port |
PUBLIC_URL |
http://localhost:3001 |
Base URL returned in browser review links |
MCP_ENDPOINT |
/mcp |
Streamable HTTP MCP path |
MCP_STATEFUL |
true |
Enables stateful MCP sessions |
MCP_JSON_RESPONSE |
false |
Returns JSON responses instead of opening SSE streams |
DB_PATH |
/app/data/steps.db in Docker |
SQLite database path |
ALLOWED_ORIGINS |
empty | Additional comma-separated browser origins |
Local development
Requires Node.js 20 or newer.
npm ci
npm run typecheck
npm test
npm run build
npm start
The default local database is steps.db. Copy .env.example to .env to customize local settings.
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。