oflow-mcp

oflow-mcp

An agent-native workflow MCP server that enables AI agents to execute text-defined, versionable workflows with checkpointing and state management.

Category
访问服务器

README

oflow-mcp

Agent-native workflow kernel. 工作流不必只能是 Dify、n8n 或扣子。

oflow-mcp is a workflow-only MCP server. It treats workflow as an open execution protocol for AI Agents: text-defined, versionable, checkpointed, recoverable, and callable through MCP tools.

Product positioning

Traditional workflow platforms often center on visual canvases, proprietary node graphs, and hosted platform state. oflow-mcp starts from a different premise:

  • Agent native: prompts, outputs, checkpoints, and step state are first-class workflow concepts.
  • Text is the source of truth: workflows are flow.yaml + prompts/*.md, so they can be reviewed, diffed, versioned, and reused.
  • Verifiable execution: each step can require outputs, natural confirmations, deterministic checks, and persisted state.
  • Local-first kernel: the first version runs on MCP + filesystem; UI, connectors, triggers, remote execution, and enterprise governance can layer on top later.
  • Replacement path, not a plugin: the long-term goal is to replace the core capabilities of general workflow tools such as Dify, n8n, and Coze/扣子, starting with the execution kernel.

Non-goals for the first release

This first release intentionally excludes:

  • TAPD, Confluence, GitLab, CI, or IM integrations
  • memory, inbox, init, or instructions tools from flow-mcp
  • visual canvas UI
  • database storage
  • multi-tenant permissions

Install

npm install
npm run build

Start

npm start

MCP configuration example:

{
  "mcpServers": {
    "oflow-mcp": {
      "command": "node",
      "args": ["/path/to/oflow-mcp/dist/index.js"],
      "env": {
        "OFLOW_MCP_FLOWS_DIR": "/path/to/oflow-mcp/flows",
        "OFLOW_MCP_DATA_DIR": "/tmp/oflow-mcp-instances"
      }
    }
  }
}

Environment variables

Variable Default Description
OFLOW_MCP_HOME ~/.oflow-mcp Base data directory
OFLOW_MCP_FLOWS_DIR $OFLOW_MCP_HOME/flows Workflow template directory
OFLOW_MCP_DATA_DIR $OFLOW_MCP_HOME/instances Workflow instance directory

Tools

oflow-mcp exposes only workflow tools:

Tool Description
workflow_list_templates List available templates
workflow_get_template Get template details
workflow_start Start a workflow instance
workflow_current Get current step and rendered prompt
workflow_advance Complete current step and advance
workflow_status Show full instance status
workflow_list_instances List instances
workflow_bind Bind alias to an instance
workflow_override_prompt Override one step prompt for one instance
workflow_create_template Create a template from YAML-like data and prompts

No flow_memory_*, flow_inbox_*, flow_init, TAPD, or Confluence tools are exposed.

Template structure

flows/
  basic-dev/
    flow.yaml
    prompts/
      analyze.md
      design.md
      verify.md

Minimal flow.yaml:

name: basic-dev
description: Minimal Agent-native development workflow
params:
  change_name:
    type: string
    required: true
steps:
  - id: analyze
    name: Analyze
    checkpoint:
      required_outputs:
        analysis_summary:
          type: string
          min_length: 20
      conditions:
        - natural: analysis_summary has been produced
          check: outputs.analysis_summary != null AND len(outputs.analysis_summary) > 20
    next: design
  - id: design
    name: Design
    next: null

Prompt variables:

  • {{change_name}} reads workflow params.
  • {{steps.analyze.outputs.analysis_summary}} reads prior step outputs.
  • Unresolved variables are left unchanged for debugging.

DSL support matrix

Feature Status
params object and string-array compatibility Supported
steps with id, name, checkpoint, next Supported
next as string/null/object branch map Supported
prompts/<step_id>.md Supported
required_outputs array or object Supported
natural conditions Supported
deterministic check expressions Supported subset
token_budget.total and token_consumed Supported
loops Not supported in first release
optimization hints Not supported
worklog hooks Not supported
inbox/memory/external bindings Not supported

Supported check expressions:

  • outputs.foo != null
  • outputs.foo == null
  • outputs.foo == 'value'
  • len(outputs.foo) > N
  • AND, OR, parentheses

Unsupported expressions fail closed and do not mutate workflow state.

Example lifecycle

  1. workflow_list_templates
  2. workflow_start:
{
  "template": "basic-dev",
  "params": { "change_name": "demo" },
  "alias": "demo-run"
}
  1. workflow_current with demo-run
  2. workflow_advance with required outputs and confirmed conditions
  3. workflow_status
  4. Continue workflow_advance until completed

Development

npm install
npm run build
npm test

Common errors

  • Template not found: set OFLOW_MCP_FLOWS_DIR or copy templates to ~/.oflow-mcp/flows.
  • Prompt not found: every step requires prompts/<step_id>.md.
  • Checkpoint validation failed: provide required outputs and confirmed conditions.
  • No branch matched: pass a condition_result matching the branch keys in next.
  • Alias already bound: choose another alias or use the existing instance ID.

推荐服务器

Baidu Map

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。

官方
精选
JavaScript
Playwright MCP Server

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。

官方
精选
TypeScript
Magic Component Platform (MCP)

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。

官方
精选
本地
TypeScript
Audiense Insights MCP Server

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。

官方
精选
本地
TypeScript
VeyraX

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。

官方
精选
本地
graphlit-mcp-server

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。

官方
精选
TypeScript
Kagi MCP Server

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。

官方
精选
Python
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选
Neon MCP Server

Neon MCP Server

用于与 Neon 管理 API 和数据库交互的 MCP 服务器

官方
精选
Exa MCP Server

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。

官方
精选