MCP Graph Workflow: A Local-First Tool for Converting Product Requirement Documents into Executable Task Graphs

MCP Graph Workflow is a local-first CLI tool based on the MCP protocol. Its core function is to automatically convert Product Requirement Documents (PRDs) into structured executable task graphs. It supports AI intelligent context management, semantic search, and multi-agent orchestration, providing AI engineering teams with a complete project planning solution. The tool uses a local-first architecture, storing data in SQLite to ensure privacy and offline availability, while deeply integrating with mainstream AI coding assistants.

With the widespread application of Large Language Models (LLMs) in software development, AI engineering teams face the challenge of converting complex PRDs into executable and traceable task flows. Traditional project management tools require extensive manual configuration, while AI-assisted development needs more flexible and intelligent context management mechanisms. MCP Graph Workflow was born in this context, connecting PRDs and execution tasks via the MCP protocol to provide a local-first AI-driven development process solution.

PRD to Task Graph Conversion

Supports importing PRDs in Markdown, TXT, PDF, and HTML formats. Uses natural language processing technology to extract requirement points and build hierarchical task trees, including fields for description, priority, dependency relationships, and status tracking.

AI Context Optimization

Adopts a three-level compression strategy (summary/standard/depth), reducing token usage by 70-85% to balance the issues of information overload and insufficiency.

Intelligent Task Routing

The built-in next command recommends the optimal task to execute based on priority, dependency relationships, and blocking status, adapting to multi-threaded development scenarios.

Semantic Search and RAG

Integrates BM25 full-text search and TF-IDF vector embedding technologies, completing retrieval locally to quickly locate historical decisions, technical solutions, and code snippets.

Multi-Agent Grid Architecture

Coordinates agents such as Serena (code analysis), GitNexus (Git graph), Context7 (document retrieval), and Playwright (browser verification) to work collaboratively via an event bus, following the MCP protocol to be compatible with mainstream AI tools.

Web Dashboard Features

Provides 6 core tabs:

• Graph View: Interactive task dependency graph (supports filtering/zooming)
• PRD & Backlog: Requirement progress tracking and backlog management
• GitNexus: Intelligent code analysis (dependency relationships/impact scope)
• Serena: Code memory bank (records key design decisions)
• Insights: Bottleneck identification and efficiency metric analysis
• Benchmark: Token economy evaluation

Application Scenarios

• AI Engineers: Structured agent workflows and efficient context management
• Technical Leaders: Automatic decomposition of PRDs into traceable task graphs
• Independent Developers: AI-driven project planning and progress tracking
• Copilot/Claude/Cursor Teams: Native MCP tool integration

Technical Specifications

• 32 MCP tools, 44 REST endpoints (distributed across 17 routers), 6 CLI commands
• Test Coverage: 910+ test cases (101 Vitest unit tests, 11 Playwright end-to-end tests)

MCP Graph Workflow represents an important direction for AI-native development tools: designing workflows around AI capability models, standardizing interaction interfaces via the MCP protocol, ensuring data sovereignty and offline availability with a local-first architecture, and achieving an intelligent development experience through multi-agent collaboration. For teams exploring AI-driven development processes, it is an open-source project worth paying attention to and trying.