flight505-mcp_dincoder

Driven Intent Negotiation — Contract-Oriented Deterministic Executable Runtime

The MCP implementation of GitHub's Spec Kit methodology — transforming specifications into executable artifacts

Table of Contents

• What is DinCoder?
• Installation
• Quickstart
• MCP Prompts (AI Workflow Orchestration)
• Complete Workflow
• Available Tools
• Examples
• Why Spec-Driven Development?
• Roadmap
• Contributing

🎯 What is DinCoder?

An official Model Context Protocol server implementing GitHub's Spec-Driven Development (SDD) methodology

DinCoder brings the power of GitHub Spec Kit to any AI coding agent through the Model Context Protocol. It transforms the traditional "prompt-then-code-dump" workflow into a systematic, specification-driven process where specifications don't serve code—code serves specifications.

What's New in v0.4.0 (Integration & Discovery Update)

🎯 MCP Prompts - AI Workflow Orchestration ✨

• 7 workflow prompts that guide AI agents through complex tasks
• Automatic discovery: AI agents find and use prompts programmatically
• Built-in guidance: Each prompt includes comprehensive workflow instructions
• Works everywhere: Claude Code, VS Code Copilot, OpenAI Codex, Cursor
• Natural language: Just describe what you want - AI uses appropriate prompts automatically

Available Prompts:

• start_project - Initialize new spec-driven project
• create_spec - Create feature specification
• generate_plan - Generate implementation plan
• create_tasks - Break down into actionable tasks
• review_progress - Generate progress report
• validate_spec - Check specification quality
• next_tasks - Show actionable tasks

Note: These are NOT slash commands you type. They're workflow templates that your AI agent uses automatically when you describe your goals!

🧬 Constitution Tool - Define Your Project's DNA

• New command: constitution_create
• Set project-wide principles, constraints, and preferences
• Ensures consistency across all AI-generated code

❓ Clarification Tracking - Systematic Q&A Management

• New commands: clarify_add, clarify_resolve, clarify_list
• Track ambiguities with unique IDs (CLARIFY-001, CLARIFY-002, etc.)
• Resolve uncertainties with rationale and audit trail

📦 Installation

🎯 Quick Decision Guide:

• Using Claude Code? → Install the Plugin (easier, includes slash commands & agents)
• Using VS Code/Codex/Cursor? → Install MCP Server Only (plugins not supported)

⚠️ Don't install both! The plugin automatically installs the MCP server - installing both may cause conflicts.

Prerequisites

• Node.js >= 20.0.0
• npm or pnpm
• An MCP-compatible coding assistant with automatic workspace binding (Cursor, Claude Code, Codex, etc.)

Installing via Smithery

To install DinCoder automatically via Smithery:

npx -y @smithery/cli install @flight505/mcp_dincoder

Claude Code / VS Code Users

claude mcp add dincoder -- npx -y mcp-dincoder@latest

Cursor

Configure the MCP server inside Cursor's MCP settings; once you select a project, Cursor injects the workspace path automatically.

Other MCP Clients

Install globally:

npm install -g mcp-dincoder@latest

Recommended clients: DinCoder expects the MCP client to bind the active project directory automatically so generated specs, plans, and tasks land in the repo you are working on. Cursor, Claude Code, and Codex do this for every request. Claude Desktop's chat UI does not, so commands default to the server's own install directory; only use Claude Desktop if you plan to pass workspacePath manually on each call.

📁 Where Files Are Created

Important: DinCoder creates all files in your current working directory (where you run your AI agent from).

your-project/
├── specs/                    # Created automatically
│   ├── 001-feature-name/    # Feature directory (auto-numbered)
│   │   ├── constitution.md  # Project principles (optional, recommended first step)
│   │   ├── spec.md          # Requirements & user stories
│   │   ├── plan.md          # Technical implementation plan
│   │   ├── tasks.md         # Executable task list
│   │   ├── research.md      # Technical decisions & research
│   │   ├── clarifications.json  # Q&A tracking
│   │   └── contracts/       # API contracts, data models
│   └── 002-next-feature/
└── .dincoder/               # Backward compatibility (legacy)

Tip: Launch your MCP client from the project root so every tool writes into the correct repo.

🚀 Quickstart

The Spec Kit Workflow

Transform ideas into production-ready code through three powerful commands:

1️⃣ /specify — Transform Ideas into Specifications

/specify Build a team productivity platform with Kanban boards and real-time collaboration

What happens:

• Automatic feature numbering (001, 002, 003...)
• Branch creation with semantic names
• Template-based specification generation
• Structured requirements with user stories
• Explicit uncertainty markers [NEEDS CLARIFICATION]

Output: A comprehensive PRD focusing on WHAT users need and WHY—never HOW to implement.

2️⃣ /plan — Map Specifications to Technical Decisions

/plan Use Next.js with Prisma and PostgreSQL, WebSockets for real-time updates

What happens:

• Analyzes feature specification
• Ensures constitutional compliance (architectural principles)
• Translates requirements to technical architecture
• Generates data models, API contracts, test scenarios
• Documents technology rationale

Output: Complete implementation plan with every decision traced to requirements.

3️⃣ /tasks — Generate Executable Task Lists

/tasks

What happens:

• Analyzes plan and contracts
• Converts specifications into granular tasks
• Marks parallelizable work [P]
• Orders tasks by dependencies
• Creates test-first implementation sequence

Output: Numbered task list ready for systematic implementation.

Real-World Example: Building a Chat System

See how SDD transforms traditional development (click to expand)

Traditional Approach (12+ hours of documentation)

1. Write PRD in document (2-3 hours)
2. Create design documents (2-3 hours)
3. Set up project structure (30 minutes)
4. Write technical specs (3-4 hours)
5. Create test plans (2 hours)

SDD with DinCoder (15 minutes total)

# Step 1: Create specification (5 minutes)
/specify Real-time chat with message history, user presence, and typing indicators

# Automatically creates:
# - Branch "003-real-time-chat"
# - specs/003-real-time-chat/spec.md with:
#   • User stories and personas
#   • Acceptance criteria
#   • [NEEDS CLARIFICATION] markers for ambiguities

# Step 2: Generate implementation plan (5 minutes)
/plan WebSocket for real-time, PostgreSQL for history, Redis for presence

# Generates:
# - plan.md with phased implementation
# - data-model.md (Message, User, Channel schemas)
# - contracts/websocket-events.json
# - contracts/rest-api.yaml
# - research.md with library comparisons

# Step 3: Create task list (5 minutes)
/tasks

# Produces executable tasks:
# 1. [P] Create WebSocket contract tests
# 2. [P] Create REST API contract tests
# 3. Set up PostgreSQL schema
# 4. Implement message persistence
# 5. Add Redis presence tracking
# ... (numbered, ordered, parallelizable)

Result: Complete, executable specifications ready for any AI agent to implement.

🎯 MCP Prompts (AI Workflow Orchestration)

New in v0.4.0: DinCoder includes 7 MCP prompts that provide guided workflows for AI agents. These are NOT slash commands you type—they're workflow templates that your AI agent (Claude, Copilot, etc.) automatically discovers and uses to help you.

How MCP Prompts Work

MCP prompts are invisible to users but powerful for AI agents:

1. AI Discovery: When DinCoder is connected, your AI agent automatically discovers available prompts via the MCP protocol
2. AI Invocation: The AI agent invokes prompts programmatically when they're relevant to your task
3. Workflow Guidance: Each prompt includes comprehensive instructions for multi-step workflows
4. Tool Orchestration: Prompts guide the AI to call multiple DinCoder tools in the correct sequence

You don't "run" these prompts directly. Just describe what you want in natural language, and your AI agent will use the appropriate prompt workflow automatically!

Available Workflow Prompts

Example: How Prompts Guide AI Workflows

You say: "Let's start a new task manager project"

AI thinks: This matches the start_project prompt. Let me follow its workflow...

AI does:

1. Calls specify_start tool with projectName="task-manager"
2. Explains the .dincoder/ structure created
3. Asks what you want to build
4. Calls specify_describe with your requirements
5. Validates spec with spec_validate
6. Suggests next steps

You don't see: The prompt invocation—just the AI following the workflow naturally!

See detailed prompt workflows (click to expand)

1. start_project - Initialize New Spec-Driven Project

AI receives this workflow when you want to start a project:

1. Call `specify_start` with projectName and agent type
2. Explain .dincoder/ directory structure to user
3. Explain spec-driven workflow: Specify → Plan → Execute
4. Ask what they want to build
5. Guide through specification creation

Example conversation:

• You: "I want to start a new e-commerce project"
• AI: Invokes start_project prompt, follows workflow
• AI: "I'll initialize a new spec-driven project. What features should the e-commerce platform have?"

2. create_spec - Create Feature Specification

AI receives this workflow when you describe a feature:

1. Check if .dincoder/ exists (run specify_start if not)
2. Gather requirements by asking:
   - What problem does this solve?
   - Who are the users?
   - What are success criteria?
   - What's out of scope?
3. Call `specify_describe` with complete specification
4. Call `spec_validate` to check quality
5. Address validation issues with `spec_refine`
6. Confirm spec is complete

Example conversation:

• You: "Build a real-time chat feature with typing indicators"
• AI: Invokes create_spec prompt, asks clarifying questions
• AI: "Let me create a specification. Should the chat support file attachments?"

3. generate_plan - Generate Implementation Plan

AI receives this workflow when planning is needed:

1. Verify spec exists (guide user to create if missing)
2. Run `spec_validate` if not already validated
3. Call `plan_create` with technical constraints
4. Call `artifacts_analyze` to verify spec-plan alignment
5. Present plan structure to user
6. Ask if ready for task generation

Example conversation:

• You: "How should we implement this?"
• AI: Invokes generate_plan prompt
• AI: "I'll create a technical plan. What's your preferred tech stack? (Next.js, Python/FastAPI, etc.)"

4. create_tasks - Break Down into Actionable Tasks

AI receives this workflow for task generation:

1. Verify plan exists (guide to create if missing)
2. Call `tasks_generate` with granular scope
3. Call `tasks_visualize` to show dependency graph
4. Call `tasks_stats` to show effort estimates
5. Call `tasks_filter` with preset:"next" for actionable items
6. Guide user to start first task

Example conversation:

• You: "Let's break this down into tasks"
• AI: Invokes create_tasks prompt
• AI: "I've generated 23 tasks. Here's the dependency graph... Ready to start with T001?"

5. review_progress - Generate Progress Report

AI receives this workflow for progress checks:

1. Call `tasks_stats` with charts and blockers
2. Call `tasks_filter` preset:"next" for actionable items
3. Call `tasks_search` for in-progress tasks
4. Summarize in structured format:
   - Overall completion %
   - Recent accomplishments
   - Current work
   - Next actions (unblocked)
   - Blockers
   - Recommendations
5. Include visual progress charts

Example conversation:

• You: "How's the project coming along?"
• AI: Invokes review_progress prompt
• AI: "We're 67% complete (15/23 tasks). 3 tasks in progress, 2 blocked. Next up: authentication tests."

6. validate_spec - Check Specification Quality

AI receives this workflow for validation:

1. Call `spec_validate` with all checks enabled
2. Check for:
   - Completeness (all required sections)
   - Acceptance criteria (testable when/then)
   - Clarifications (no unresolved markers)
   - Implementation leakage (no HOW in WHAT)
3. Call `clarify_list` for pending questions
4. If validation fails:
   - List specific issues
   - Call `spec_refine` to fix
   - Re-validate
5. If passes: suggest `plan_create` next

Example conversation:

• You: "Is the spec ready for implementation?"
• AI: Invokes validate_spec prompt
• AI: "Validation found 2 issues: missing edge cases section, 1 unresolved clarification. Let me fix these..."

7. next_tasks - Show Next Actionable Tasks

AI receives this workflow when you ask what's next:

1. Call `tasks_filter` with preset:"next" and priority sorting
2. For each task display:
   - Task ID and description
   - Metadata (phase, type, priority, effort)
   - Why it's actionable
3. Recommend which task to start based on:
   - Priority level
   - Dependencies completed
   - Effort estimate
4. Offer task details on request

Example conversation:

• You: "What should I work on next?"
• AI: Invokes next_tasks prompt
• AI: "Top priority: T007 (Implement user authentication, effort: 5). It's unblocked and high priority. Want to start?"

Platform Compatibility

MCP prompts work across all MCP-compatible clients:

The Key Difference: MCP Prompts vs Slash Commands

Important distinction:

• MCP Prompts (DinCoder workflows): AI agents use these programmatically. You don't type them.
• Slash Commands (Native): User-typed commands like /help, /clear in Claude Code
• Custom Commands (.claude/commands/): Project-specific slash commands you create

In practice: You describe what you want in natural language ("Let's start a new project"), and your AI agent automatically uses the appropriate MCP prompt workflow!

🔌 Claude Code Plugin (Recommended for Claude Code)

New in v0.5.0: For the best Claude Code experience, install the DinCoder Plugin which bundles slash commands, specialized agents, and automatically installs the MCP server.

⚠️ Important: The plugin includes the MCP server - you don't need to install both! Choose one installation method:

• Plugin (Claude Code only) → Slash commands + agents + MCP server (all-in-one)
• MCP Server only (VS Code, Codex, etc.) → Just the tools (manual setup required)

Prerequisites

Before installing the plugin:

• ✅ Claude Code version 2.0.13 or higher
• ✅ Node.js >= 18
• ✅ npm installed

Installation

Step 1: Add the DinCoder marketplace

# In Claude Code
/plugin marketplace add flight505/dincoder-plugin

Step 2: Install the plugin

/plugin install dincoder

Step 3: Restart Claude Code

• Press Cmd+Q (Mac) or Alt+F4 (Windows) to quit, then reopen
• Or use Cmd/Ctrl+Shift+P → "Developer: Reload Window"

Verify Installation

After restart, check that:

• Slash commands appear: /spec, /plan, /tasks, /progress, /validate, /next
• Agents appear: @spec-writer, @plan-architect, @task-manager
• MCP server is active in Settings → Extensions → MCP Servers

What's Included

✨ Slash Commands - Quick access without memorizing tool names

• /spec - Create or refine specification
• /plan - Generate implementation plan
• /tasks - Break down into actionable tasks
• /progress - View progress report
• /validate - Check spec quality
• /next - Show next actionable tasks

🤖 Specialized Agents - Expert assistance for each phase

• @spec-writer - Expert at creating validated specifications
• @plan-architect - Expert at designing technical plans
• @task-manager - Expert at managing tasks and progress

🔧 Automatic MCP Server - Installs and configures mcp-dincoder@latest from npm automatically

• Runs npx -y mcp-dincoder@latest on installation
• Always pulls the latest version for bug fixes and features
• No manual MCP server setup needed!

📝 Built-in Documentation - CLAUDE.md loads automatically with methodology guide

Plugin vs MCP Server Only

Choose Your Installation Method:

• Claude Code users: Use the plugin (recommended) - get slash commands, agents, and MCP server in one package
• VS Code/Codex users: Use the MCP server only (plugins not supported on these platforms)

⚠️ Avoid Dual Installation: If you're using the plugin, do NOT also manually configure the MCP server in your MCP settings. The plugin handles this automatically and doing both may cause conflicts.

Plugin Repository: flight505/dincoder-plugin

🔌 VS Code + GitHub Copilot Integration

New in v0.6.0: Full support for VS Code with GitHub Copilo