aifastdb-devplan

AI-Powered Development Plan Management — MCP Server

Let AI assistants (Cursor / Claude Desktop) manage your development plans, task tracking, and project documentation.
Built on the aifastdb high-performance storage engine, seamlessly integrated with AI via the MCP protocol.

让 AI 助手（Cursor / Claude Desktop）直接管理你的开发计划、任务追踪和项目文档。
基于 aifastdb 高性能存储引擎，通过 MCP 协议 与 AI 无缝集成。

English | 中文

---

English

Why aifastdb-devplan?

In the age of AI-assisted programming, developers collaborate with AI assistants more closely than ever. However, AI assistants lack persistent project memory — each conversation starts from scratch, with no knowledge of the overall project plan, current progress, or historical decisions.

aifastdb-devplan solves this problem by providing AI assistants with a set of structured development plan management tools, enabling AI to:

• 📋 Understand the big picture — Read project overviews, technical designs, API specifications, and more
• 🎯 Track task progress — Manage two-level task hierarchies (MainTask → SubTask) with real-time progress updates
• 🔗 Anchor to Git history — Automatically record Git commit hashes when completing tasks, with rollback detection
• 📦 Modular management — Organize tasks and docs by feature modules for a clear project architecture view
• 📄 Export documentation — Generate complete Markdown-formatted development plan documents in one click

Key Features

Feature	Description
Dual Storage Engine	Choose between graph (SocialGraphV2, default) or document (JSONL) per project
Project Graph Page	Built-in HTTP server + vis-network page to visualize tasks/modules as an interactive graph
11 Document Section Types	overview, requirements, api_design, technical_notes, architecture, and more
Two-Level Task Hierarchy	MainTask + SubTask with priority levels (P0–P3) and status transitions
Module Registry	Aggregate tasks and docs by module for intuitive project architecture
Git Commit Anchoring	Auto-record commit hash on task completion; sync_git detects rollbacks
Auto Progress Tracking	Automatically update parent task progress when subtasks are completed
Idempotent Task Import	upsert_task prevents duplicates, ideal for batch initialization
Data Migration	Seamlessly migrate between document and graph engines with backup support
Markdown Export	Generate structured development plan documents for sharing and archiving
Zero-Config Storage	Local storage in the project's .devplan/ directory — no external database needed

Quick Start

Installation

npm install -g aifastdb-devplan

Option A: As an MCP Server (Recommended)

Configure in Cursor IDE (.cursor/mcp.json):

{
  "mcpServers": {
    "aifastdb-devplan": {
      "command": "npx",
      "args": ["aifastdb-devplan"]
    }
  }
}

Or in Claude Desktop (claude_desktop_config.json):

{
  "mcpServers": {
    "aifastdb-devplan": {
      "command": "npx",
      "args": ["aifastdb-devplan"]
    }
  }
}

Once configured, your AI assistant can use the default micro MCP tool set (18 core devplan_* tools) to manage your development plans.

MCP Tool Exposure Modes

aifastdb-devplan now supports three MCP tool exposure modes:

• micro (default): expose 18 core tools for the most common DevPlan workflow
• slim: expose 30 commonly used tools
• full: expose the complete MCP tool catalog for advanced or low-frequency workflows

The exposed tools are organized into five groups:

• project
• docs
• tasks
• memory
• batch

By default, the server starts in micro mode, which means:

• MCP ListTools only returns those 18 tools
• only those 18 tools are exposed to the AI assistant
• tools outside that set are not just hidden; they are rejected at call time

To temporarily enable the full catalog:

export AIFASTDB_DEVPLAN_MCP_TOOL_MODE=full
npx aifastdb-devplan

PowerShell:

$env:AIFASTDB_DEVPLAN_MCP_TOOL_MODE = "full"
npx aifastdb-devplan

To explicitly use slim mode:

export AIFASTDB_DEVPLAN_MCP_TOOL_MODE=slim

PowerShell:

$env:AIFASTDB_DEVPLAN_MCP_TOOL_MODE = "slim"

To explicitly use micro mode:

export AIFASTDB_DEVPLAN_MCP_TOOL_MODE=micro

PowerShell:

$env:AIFASTDB_DEVPLAN_MCP_TOOL_MODE = "micro"

Option B: As an npm Package (Programmatic)

import { DevPlanStore, createDevPlan } from 'aifastdb-devplan';

const plan = createDevPlan('my-project');

// Create a main task
plan.createMainTask({
  projectName: 'my-project',
  taskId: 'phase-1',
  title: 'Phase 1: Foundation Setup',
  priority: 'P0',
});

// Add a subtask
plan.addSubTask({
  projectName: 'my-project',
  taskId: 'T1.1',
  parentTaskId: 'phase-1',
  title: 'Initialize project structure',
});

// Complete task (auto-updates progress + Git commit anchoring)
plan.completeSubTask('T1.1');

// Check progress
const progress = plan.getProgress();
console.log(progress);

MCP Tools (default micro mode: 18 total, slim mode: 30 total)

Grouped Catalog

Group	Tools
project	devplan_init, devplan_save_prompt, devplan_get_progress
docs	devplan_save_section, devplan_get_section, devplan_list_sections, devplan_search_sections, devplan_delete_section
tasks	devplan_create_main_task, devplan_add_sub_task, devplan_upsert_task, devplan_delete_task, devplan_update_task_status, devplan_complete_task, devplan_list_tasks, devplan_search_tasks, devplan_start_phase
memory	devplan_memory_save, devplan_recall_unified, devplan_memory_context, devplan_memory_list, devplan_memory_delete, devplan_memory_generate
batch	devplan_memory_batch_prepare, devplan_memory_batch_commit, devplan_memory_batch_status
modules	devplan_create_module, devplan_list_modules, devplan_get_module, devplan_update_module

Default micro Mode

micro keeps the shortest, highest-frequency workflow tools:

• project: devplan_init, devplan_save_prompt
• docs: devplan_get_section, devplan_search_sections, devplan_save_section
• tasks: devplan_list_tasks, devplan_search_tasks, devplan_start_phase, devplan_create_main_task, devplan_add_sub_task, devplan_upsert_task, devplan_complete_task
• memory: devplan_memory_save, devplan_recall_unified
• modules: devplan_create_module, devplan_list_modules, devplan_get_module, devplan_update_module

slim expands that to the 30 commonly used tools above, and full exposes the entire catalog.

Dual Storage Engine

Each project can independently choose its storage engine:

Engine	Backend	Default	Features
graph	SocialGraphV2 (WAL + sharding)	✅ New projects	Graph visualization, entity-relation model
document	DocumentStore (JSONL)	Auto-detected for legacy	Lightweight, human-readable files

Engine selection priority:

1. Explicit engine parameter in createDevPlan()
2. .devplan/{project}/engine.json configuration
3. Auto-detect existing JSONL files → document
4. New projects → graph

Project Graph Page

Visualize your development plan as an interactive graph:

npm run visualize -- --project my-project --base-path /path/to/.devplan
# or
aifastdb-devplan-visual --project my-project --port 3210

The built-in HTTP server serves a self-contained HTML page with vis-network, featuring:

• 5 node types: project (star), module (diamond), main-task (circle), sub-task (dot), document (box)
• Status-based coloring: completed (green), in-progress (blue), pending (gray)
• Interactive features: click for details panel, filter by type, stats bar with progress
• Dark theme: consistent with modern development tools

Enable DevPlan in Other Projects (Step-by-Step Guide)

Here's a complete guide to enable devplan in any project (e.g., my-app).

Method 1: npm Published Version (Recommended)

Step 1: Install globally

npm install -g aifastdb-devplan

Step 2: Configure MCP Server in your project

Create .cursor/mcp.json in your project root:

{
  "mcpServers": {
    "aifastdb-devplan": {
      "command": "npx",
      "args": ["aifastdb-devplan"]
    }
  }
}

This npx form is the portable default for npm-installed usage because it relies on the package bin entry that npm generates for the current machine.

If you hit a Cursor reload/reconnect issue where a split command + args startup can be resumed incompletely, prefer wrapping the full startup command in a single launcher script and point command to that script instead of node:

Windows launcher example:

@echo off
"C:\Program Files\nodejs\node.exe" "D:\Project\git\aifastdb-devplan\dist\mcp-server\index.js" %*

Then use:

{
  "mcpServers": {
    "aifastdb-devplan": {
      "command": "D:\\Project\\git\\aifastdb-devplan\\scripts\\start-mcp.cmd"
    }
  }
}

This is the most stable Cursor configuration for local development because Cursor only has to remember one launcher path, not a separate node command plus script args.

Step 3: Start using with AI assistant

Open Cursor in your project directory and tell the AI:

Initialize a development plan for my-app project

The AI will call devplan_init and data will be stored in .devplan/my-app/ under your project root (auto-detected via .git or package.json).

If the project already has .cursor/rules/dev-plan-management.mdc and you want to upgrade it to the latest generated template, call:

devplan_init({
  projectName: "my-app",
  refreshCursorRule: true
})

For document search, prefer explicit field targeting when possible:

devplan_search_sections({ projectName: "my-app", query: "171e9a18-c7e9-430b-9e3d-fa6d384a0b4e", searchBy: "id" })
devplan_search_sections({ projectName: "my-app", query: "Vector Store", searchBy: "title" })
devplan_search_sections({ projectName: "my-app", query: "BM25 tokenization", searchBy: "content" })

For task search, you can also explicitly control which field to search:

devplan_search_tasks({ projectName: "my-app", query: "phase-14", searchBy: "taskId" })
devplan_search_tasks({ projectName: "my-app", query: "Vector Search Refactor", searchBy: "title" })
devplan_search_tasks({ projectName: "my-app", query: "refresh cursor rule template", searchBy: "description" })
devplan_search_tasks({ projectName: "my-app", query: "rebuild search ranking helper", searchBy: "subTask" })
devplan_search_tasks({ projectName: "my-app", query: "search ranking", includeSubTasks: true })

When a sub-task matches, matchedSubTasks returns the matched child items. If includeSubTasks: true, the response also includes all sub-tasks under each matched main task.

Method 2: Local Development Version

If you're working with a local clone of aifastdb-devplan (not yet published or testing changes):

Step 1: Build locally

cd /path/to/aifastdb-devplan
npm install
npm run build

Step 2: Configure MCP Server with local path

Create .cursor/mcp.json in your target project:

{
  "mcpServers": {
    "aifastdb-devplan": {
      "command": "/path/to/aifastdb-devplan/scripts/start-mcp.sh"
    }
  }
}

Windows example:

{
  "mcpServers": {
    "aifastdb-devplan": {
      "command": "D:\\Project\\git\\aifastdb-devplan\\scripts\\start-mcp.cmd"
    }
  }
}

Recommended local launcher scripts:

Unix-like:

#!/usr/bin/env bash
node "/path/to/aifastdb-devplan/dist/mcp-server/index.js" "$@"

Windows:

@echo off
"D:\Program Files\nodejs\node.exe" "D:\Project\git\aifastdb-devplan\dist\mcp-server\index.js" %*

Use the direct node + args form only if you specifically want the simplest config and have not seen reload-related startup truncation in Cursor.

Controlling Data Storage Location

By default, devplan auto-detects your project root and stores data in .devplan/. You can override this:

Option A: Environment variable (global override)

# All devplan data will be stored under this path
export AIFASTDB_DEVPLAN_PATH=/path/to/shared/devplans

Option B: --base-path for visualization server

# View another project's devplan graph
aifastdb-devplan-visual --project my-app --base-path /path/to/my-app/.devplan --port 3210

--base-path Parameter Details

Parameter	Description	Default
--project	Project name (must match the name used in devplan_init)	Required
--base-path	Absolute path to the .devplan directory	Auto-detect via .git / package.json, fallback to ~/.aifastdb/dev-plans/
--port	HTTP server port	3210

Data directory structure under --base-path:

<base-path>/
└── <project-name>/
    ├── engine.json        # Engine config
    ├── graph-data/        # Graph engine data (WAL shards)
    ├── documents.jsonl    # Document engine data
    ├── tasks.jsonl
    └── modules.jsonl

Complete Example: Managing "my-app" from Scratch

# 1. Install devplan globally
npm install -g aifastdb-devplan

# 2. Go to your project
cd /path/to/my-app

# 3. Create MCP config
mkdir -p .cursor
echo '{"mcpServers":{"aifastdb-devplan":{"command":"npx","args":["aifastdb-devplan"]}}}' > .cursor/mcp.json

# 4. Open in Cursor and tell AI:
#    "Initialize devplan for my-app, create Phase 1 with 3 subtasks"

# 5. Visualize the plan graph
npx aifastdb-devplan-visual --project my-app --base-path .devplan --port 3210

Data Storage

Data is stored locally — no external database required:

.devplan/{projectName}/
├── engine.json        # Engine configuration (graph or document)
├── documents.jsonl    # Document sections (document engine)
├── tasks.jsonl        # Main tasks + subtasks (document engine)
├── modules.jsonl      # Feature modules (document engine)
└── graph-data/        # WAL-based graph storage (graph engine)
    └── wal/           # Write-ahead log shards

Storage path resolution priority:

Priority	Source	Description
1	AIFASTDB_DEVPLAN_PATH env var	Explicitly specify storage directory
2	.devplan/ in project root	Auto-detect via .git / package.json
3	~/.aifastdb/dev-plans/	Global fallback path

Platform Support

aifastdb-devplan is a pure TypeScript/JavaScript project, supporting all platforms with Node.js ≥ 18:

Platform	Architecture	Supported
Windows	x64	✅
macOS	x64 / Apple Silicon (M1/M2/M3/M4)	✅
Linux	x64 / ARM64	✅

Note: The underlying storage engine aifastdb includes Rust native bindings with prebuilt binaries for all listed platforms.

---

中文

为什么需要 aifastdb-devplan？

在 AI 辅助编程时代，开发者与 AI 助手的协作越来越密切。但 AI 助手缺乏持久化的项目记忆——每次对话都从零开始，无法了解项目的整体规划、当前进度和历史决策。

aifastdb-devplan 解决了这个问题：它为 AI 助手提供了一套结构化的开发计划管理工具，让 AI 能够：

• 📋 了解项目全貌 — 读取项目概述、技术方案、API 设计等文档片段
• 🎯 追踪任务进度 — 管理两级任务层级（主任务 → 子任务），实时更新进度
• 🔗 锚定 Git 历史 — 完成任务时自动记录 Git commit hash，支持回滚检测
• 📦 模块化管理 — 按功能模块组织任务和文档，清晰展示项目架构
• 📄 导出文档 — 一键生成完整的 Markdown 格式开发计划文档

核心特性

特性	说明
双存储引擎	每个项目可独立选择 graph（SocialGraphV2，默认）或 document（JSONL）引擎
项目图谱	内置 HTTP 服务器 + vis-network 页面，将任务/模块以交互式图谱展示
11 种文档片段	overview, requirements, api_design, technical_notes, architecture 等标准类型
两级任务层级	主任务 (MainTask) + 子任务 (SubTask)，支持优先级 (P0-P3) 和状态流转
功能模块注册表	按模块维度聚合任务和文档，直观展示项目架构
Git Commit 锚定	完成任务时自动记录 commit hash，sync_git 可检测代码回滚
自动进度统计	完成子任务时自动更新主任务进度百分比
幂等任务导入	upsert_task 支持防重复导入，适合批量初始化
数据迁移	在 document 和 graph 引擎间无缝迁移，支持备份
Markdown 导出	生成结构化的开发计划文档，方便分享和归档
零配置存储	本地存储，数据保存在项目 .devplan/ 目录中，无需外部数据库

快速开始

安装

npm install -g aifastdb-devplan

方式 A：作为 MCP Server 使用（推荐）

在 Cursor IDE 中配置 .cursor/mcp.json：

{
  "mcpServers": {
    "aifastdb-devplan": {
      "command": "npx",
      "args": ["aifastdb-devplan"]
    }
  }
}

或在 Claude Desktop 中配置 claude_desktop_config.json：

{
  "mcpServers": {
    "aifastdb-devplan": {
      "command": "npx",
      "args": ["aifastdb-devplan"]
    }
  }
}

配置完成后，AI 助手默认即可使用 micro MCP 工具集（18 个核心 devplan_* 工具）来管理你的开发计划。

MCP 工具暴露模式

aifastdb-devplan 现在支持三种 MCP 工具暴露模式：

• micro（默认）：只暴露 18 个核心工具
• slim：暴露 30 个常用 DevPlan 工具
• full：暴露完整 MCP 工具目录，适合高级或低频工作流

工具同时按 6 组组织：

• project
• docs
• tasks
• memory
• batch
• modules

默认启动时使用 micro 模式，这意味着：

• MCP ListTools 只会返回这 18 个工具
• AI 助手上下文里也只会看到并使用这 18 个工具
• 白名单外工具不只是“隐藏”，而是调用时会被直接拒绝

临时开启全量工具目录：

export AIFASTDB_DEVPLAN_MCP_TOOL_MODE=full
npx aifastdb-devplan

PowerShell：

$env:AIFASTDB_DEVPLAN_MCP_TOOL_MODE = "full"
npx aifastdb-devplan

如需显式切换到 slim 模式：

export AIFASTDB_DEVPLAN_MCP_TOOL_MODE=slim

PowerShell：

$env:AIFASTDB_DEVPLAN_MCP_TOOL_MODE = "slim"

如需显式使用默认 micro 模式：

export AIFASTDB_DEVPLAN_MCP_TOOL_MODE=micro

PowerShell：

$env:AIFASTDB_DEVPLAN_MCP_TOOL_MODE = "micro"

方式 B：作为 npm 包编程使用

import { DevPlanStore, createDevPlan } from 'aifastdb-devplan';

const plan = createDevPlan('my-project');

// 创建主任务
plan.createMainTask({
  projectName: 'my-project',
  taskId: 'phase-1',
  title: '阶段一：基础搭建',
  priority: 'P0',
});

// 添加子任务
plan.addSubTask({
  projectName: 'my-project',
  taskId: 'T1.1',
  parentTaskId: 'phase-1',
  title: '初始化项目结构',
});

// 完成任务（自动更新主任务进度 + Git commit 锚定）
plan.completeSubTask('T1.1');

// 查看进度
const progress = plan.getProgress();
console.log(progress);

MCP 工具一览（默认 micro 18 个，slim 30 个）

分组目录

分组	工具
project	devplan_init、devplan_save_prompt、devplan_get_progress
docs	devplan_save_section、devplan_get_section、devplan_list_sections、devplan_search_sections、devplan_delete_section
tasks	devplan_create_main_task、devplan_add_sub_task、devplan_upsert_task、devplan_delete_task、devplan_update_task_status、devplan_complete_task、devplan_list_tasks、devplan_search_tasks、devplan_start_phase
memory	devplan_memory_save、devplan_recall_unified、devplan_memory_context、devplan_memory_list、devplan_memory_delete、devplan_memory_generate
batch	devplan_memory_batch_prepare、devplan_memory_batch_commit、devplan_memory_batch_status
modules	devplan_create_module、devplan_list_modules、devplan_get_module、devplan_update_module

默认 micro 模式

micro 只保留最高频、最短链路的工具：

• project：devplan_init、devplan_save_prompt
• docs：devplan_get_section、devplan_search_sections、devplan_save_section
• tasks：devplan_list_tasks、devplan_search_tasks、devplan_start_phase、devplan_create_main_task、devplan_add_sub_task、devplan_upsert_task、devplan_complete_task
• memory：devplan_memory_save、devplan_recall_unified
• modules：devplan_create_module、devplan_list_modules、devplan_get_module、devplan_update_module

当前默认 micro 的 18 个工具完整清单：

• devplan_init
• devplan_save_prompt
• devplan_get_section
• devplan_search_sections
• devplan_save_section
• devplan_list_tasks
• devplan_search_tasks
• devplan_start_phase
• devplan_create_main_task
• devplan_add_sub_task
• devplan_upsert_task
• devplan_complete_task
• devplan_memory_save
• devplan_recall_unified
• devplan_create_module
• devplan_list_modules
• devplan_get_module
• devplan_update_module

slim 扩展为上面的 30 个常用工具，full 则暴露全部工具目录。

双存储引擎

每个项目可独立选择存储引擎：

引擎	后端	默认	特点
graph	SocialGraphV2（WAL + 分片）	✅ 新项目	图可视化、实体-关系模型
document	DocumentStore（JSONL）	旧项目自动检测	轻量、文件可读

引擎选择优先级：

1. createDevPlan() 显式传入 engine 参数
2. .devplan/{project}/engine.json 配置文件
3. 已有 JSONL 数据文件 → 自动识别为 document
4. 新项目 → 默认使用 graph

项目图谱

将开发计划以交互式图谱形式展示：

npm run visualize -- --project my-project --base-path /path/to/.devplan
# 或
aifastdb-devplan-visual --project my-project --port 3210

内置 HTTP 服务器提供自包含 HTML 页面，使用 vis-network 渲染：

• 5 种节点类型：项目（星形）、模块（菱形）、主任务（圆形）、子任务（小圆点）、文档（方形）
• 状态着色：已完成（绿色）、进行中（蓝色）、待开始（灰色）
• 交互功能：点击查看详情面板、按类型过滤、顶部统计栏 + 进度条
• 暗色主题：与现代开发工具风格一致

在其它项目中启用 DevPlan（实战指南）

以下是在任意项目（例如 my-app）中启用 devplan 的完整步骤。

方式一：使用 npm 发布版本（推荐）

第 1 步：全局安装

npm install -g aifastdb-devplan

第 2 步：在目标项目中配置 MCP Server

在项目根目录创建 .cursor/mcp.json：

{
  "mcpServers": {
    "aifastdb-devplan": {
      "command": "npx",
      "args": ["aifastdb-devplan"]
    }
  }
}

这种 npx 写法适合通过 npm 安装后的通用接入，因为它依赖 npm 为当前机器自动生成的 bin 启动入口。

如果你遇到 Cursor 在 reload / 自动重连后把拆分的 command + args 启动命令恢复不完整，建议把完整启动命令包进一个单独的 launcher 脚本里，然后让 command 直接指向该脚本，而不是裸 node。

Windows launcher 示例：

@echo off
"C:\Program Files\nodejs\node.exe" "D:\Project\git\aifastdb-devplan\dist\mcp-server\index.js" %*

然后在 .cursor/mcp.json 中改成：

{
  "mcpServers": {
    "aifastdb-devplan": {
      "command": "D:\\Project\\git\\aifastdb-devplan\\scripts\\start-mcp.cmd"
    }
  }
}

对于本地开发，这种“单一 launcher 路径”是目前更稳的 Cursor 配置方式，因为 Cursor 只需要记住一个启动文件，而不是分别记住 node 和脚本参数。

第 3 步：通过 AI 助手开始使用

在 Cursor 中打开你的项目目录，对 AI 说：

为 my-app 项目初始化开发计划

AI 会调用 devplan_init，数据自动存储在项目根目录下的 .devplan/my-app/ 中（通过 .git 或 package.json 自动检测项目根目录）。

如果项目已经存在 .cursor/rules/dev-plan-management.mdc，但你想把它升级到最新模板，可显式调用：

devplan_init({
  projectName: "my-app",
  refreshCursorRule: true
})

在文档搜索时，推荐尽量显式指定目标字段：

devplan_search_sections({ projectName: "my-app", query: "171e9a18-c7e9-430b-9e3d-fa6d384a0b4e", searchBy: "id" })
devplan_search_sections({ projectName: "my-app", query: "向量存储", searchBy: "title" })
devplan_search_sections({ projectName: "my-app", query: "BM25 中文分词", searchBy: "content" })

在任务搜索时，也推荐尽量显式指定目标字段：

devplan_search_tasks({ projectName: "my-app", query: "phase-14", searchBy: "taskId" })
devplan_search_tasks({ projectName: "my-app", query: "向量搜索重构", searchBy: "title" })
devplan_search_tasks({ projectName: "my-app", query: "刷新 cursor rule 模板", searchBy: "description" })
devplan_search_tasks({ projectName: "my-app", query: "重建搜索排序 helper", searchBy: "subTask" })
devplan_search_tasks({ projectName: "my-app", query: "搜索排序", includeSubTasks: true })

当命中的是子任务时，返回结果中的 matchedSubTasks 会列出命中的子任务；如果传入 includeSubTasks: true，还会额外附带该主任务下的全部子任务。

方式二：使用本地开发版本

如果你在使用本地克隆的 aifastdb-devplan（未发布到 npm 或正在测试修改）：

第 1 步：本地构建

cd /path/to/aifastdb-devplan
npm install
npm run build

第 2 步：使用本地路径配置 MCP Server

在目标项目中创建 .cursor/mcp.json：

{
  "mcpServers": {
    "aifastdb-devplan": {
      "command": "/path/to/aifastdb-devplan/scripts/start-mcp.sh"
    }
  }
}

Windows 示例：

{
  "mcpServers": {
    "aifastdb-devplan": {
      "command": "D:\\Project\\git\\aifastdb-devplan\\scripts\\start-mcp.cmd"
    }
  }
}

推荐的本地 launcher 脚本：

类 Unix：

#!/usr/bin/env bash
node "/path/to/aifastdb-devplan/dist/mcp-server/index.js" "$@"

Windows：

@echo off
"D:\Program Files\nodejs\node.exe" "D:\Project\git\aifastdb-devplan\dist\mcp-server\index.js" %*

只有在你明确想保留最简单配置、并且没有遇到 Cursor reload 后启动命令截断问题时，才建议继续使用 node + args 的拆分写法。

控制数据存储位置

默认情况下，devplan 会自动检测项目根目录并将数据存储在 .devplan/ 下。你可以通过以下方式覆盖：

方案 A：环境变量（全局覆盖）

# 所有 devplan 数据将存储在此路径下
export AIFASTDB_DEVPLAN_PATH=/path/to/shared/devplans

Windows PowerShell：

$env:AIFASTDB_DEVPLAN_PATH = "D:\shared\devplans"

方案 B：可视化服务器使用 --base-path

# 查看另一个项目的 devplan 图谱
aifastdb-devplan-visual --project my-app --base-path /path/to/my-app/.devplan --port 3210

--base-path 参数详解

参数	说明	默认值
--project	项目名称（必须与 devplan_init 时使用的名称一致）	必填
--base-path	.devplan 目录的绝对路径	自动检测（通过 .git / package.json），兜底 ~/.aifastdb/dev-plans/
--port	HTTP 服务器端口	3210

--base-path 下的数据目录结构：

<base-path>/
└── <project-name>/
    ├── engine.json        # 引擎配置
    ├── graph-data/        # Graph 引擎数据（WAL 分片）
    ├── documents.jsonl    # Document 引擎数据
    ├── tasks.jsonl
    └── modules.jsonl

完整示例：从零管理 "my-app" 项目

# 1. 全局安装 devplan
npm install -g aifastdb-devplan

# 2. 进入你的项目目录
cd /path/to/my-app

# 3. 创建 MCP 配置
mkdir -p .cursor
echo '{"mcpServers":{"aifastdb-devplan":{"command":"npx","args":["aifastdb-devplan"]}}}' > .cursor/mcp.json

# 4. 在 Cursor 中打开项目，对 AI 说：
#    "为 my-app 初始化开发计划，创建阶段一并添加 3 个子任务"

# 5. 可视化查看计划图谱
npx aifastdb-devplan-visual --project my-app --base-path .devplan --port 3210

数据存储

数据存储在本地，无需外部数据库：

.devplan/{projectName}/
├── engine.json        # 引擎配置（graph 或 document）
├── documents.jsonl    # 文档片段（document 引擎）
├── tasks.jsonl        # 主任务 + 子任务（document 引擎）
├── modules.jsonl      # 功能模块（document 引擎）
└── graph-data/        # WAL 图存储（graph 引擎）
    └── wal/           # 预写日志分片

存储路径解析优先级：

优先级	路径来源	说明
1	AIFASTDB_DEVPLAN_PATH 环境变量	显式指定存储目录
2	项目内 .devplan/ 目录	自动检测项目根目录（通过 .git / package.json）
3	~/.aifastdb/dev-plans/	全局兜底路径

平台支持

aifastdb-devplan 是纯 TypeScript/JavaScript 项目，支持所有 Node.js ≥ 18 的平台：

平台	架构	支持
Windows	x64	✅
macOS	x64 / Apple Silicon (M1/M2/M3/M4)	✅
Linux	x64 / ARM64	✅

注：底层存储引擎 aifastdb 包含 Rust 原生绑定，已为上述平台提供预编译二进制文件。

---

Tech Stack / 技术栈

• Storage Engine / 存储引擎: aifastdb — Dual engine: SocialGraphV2 (graph) + DocumentStore (JSONL), built with Rust + N-API
• Protocol / 通信协议: MCP (Model Context Protocol) — Standard protocol for AI assistant tool invocation
• Visualization / 可视化: vis-network — Interactive graph visualization (CDN, zero dependencies)
• Runtime / 运行时: Node.js ≥ 18
• Language / 语言: TypeScript (strict mode)

Related Projects / 相关项目

• aifastdb — AI-friendly high-performance database engine (vector search + semantic indexing + agent memory)
• MCP Protocol — Model Context Protocol official documentation

License

MIT