sdd-mcp

Spec-Driven Development MCP Server - An MCP server that provides spec-driven development custom slash commands as MCP tools

Features

Specification Management: Consistent workflow from spec initialization to requirements definition, design, task breakdown, and feedback integration
TDD Implementation Support: High-quality implementation support with test-first development
Design Validation: Interactive design review and quality checks
Progress Tracking: Spec status monitoring and implementation gap visualization
Steering Documents: Manage project-wide direction with steering documents
MCP Tools: 10 development support tools available from MCP clients like Claude Code
Template Engine: Flexible prompt generation with frontmatter-enabled template engine

Requirements

Runtime: Bun >= 1.0.0 (recommended) or Node.js >= 20
OS: macOS, Linux, or WSL2 (Native Windows is not supported)

Installation

Quick Start with bunx (Recommended)

bunx sdd-mcp@latest

Quick Start with npx

npx sdd-mcp@latest

Local Installation

# With Bun (Recommended)
bun install sdd-mcp

# With npm
npm install sdd-mcp

MCP Tools

This server provides the following spec-driven development tools:

Specification Management

spec-init: Initialize a new specification (starting from project description)
spec-requirements: Generate requirements definition
spec-design: Generate design document
spec-tasks: Execute task breakdown
spec-impl: Execute implementation with TDD approach
spec-status: Check specification status
spec-feedback: Capture development feedback, generate JSON/Markdown reports, and apply updates to requirements/design/tasks

Steering Documents

steering: Update steering documents
steering-custom: Create custom steering documents

Validation

validate-design: Review design quality
validate-gap: Analyze implementation gap

Usage

Start MCP Server

# With bunx (Recommended)
bunx sdd-mcp@latest

# With npx
npx sdd-mcp@latest

Using MCP Tools

You can invoke the above tools from MCP clients. Each tool:

Loads template files
Expands parameters to generate prompts
Returns with metadata (template_id, version, allowed_tools, parameters)

See individual template files (commands/*.md) for details.

Workflow Example

Here's a complete workflow for developing a new feature with spec-driven development:

1. Optional: Create Steering Documents (First Time)

Use MCP tool: steering

This creates project-wide context documents (product.md, tech.md, structure.md) in .kiro/steering/.

2. Initialize Specification

Use MCP tool: spec-init
Parameters: "Implement user authentication with JWT tokens and refresh token rotation"

This creates:

.kiro/specs/user-authentication/spec.json (metadata)
.kiro/specs/user-authentication/requirements.md (template)

3. Generate Requirements

Use MCP tool: spec-requirements
Parameters: user-authentication

AI analyzes the project and generates comprehensive requirements in requirements.md.

Human Review Required: Review and approve the requirements.

4. Generate Design

Use MCP tool: spec-design
Parameters: user-authentication

AI creates technical design document in design.md based on approved requirements.

Human Review Required: Review and approve the design.

Quick tip: Use spec-design user-authentication -y to auto-approve requirements and skip the interactive prompt.

5. Generate Implementation Tasks

Use MCP tool: spec-tasks
Parameters: user-authentication

AI breaks down the design into concrete implementation tasks in tasks.md.

Human Review Required: Review and approve the tasks.

Quick tip: Use spec-tasks user-authentication -y to auto-approve previous phases.

6. Execute Implementation with TDD

Use MCP tool: spec-impl
Parameters: user-authentication

AI implements the feature following Test-Driven Development methodology.

Optional: Specify task numbers: spec-impl user-authentication 1,2,3 to implement specific tasks.

7. Check Status

Use MCP tool: spec-status
Parameters: user-authentication

View current phase, approval status, and implementation progress.

8. Record and Apply Feedback (Any Time)

Use MCP tool: spec-feedback
Parameters: {"feature_name":"user-authentication","mode":"report"}

This produces a JSON report and Markdown summary describing the feedback items that need to be reflected in the specification set.

After the report is reviewed, apply the changes directly to the spec documents:

Use MCP tool: spec-feedback
Parameters: {"feature_name":"user-authentication","mode":"apply","report_path":"feedback/20251013-120000-user-authentication-feedback.json"}

The apply prompt guides you to update requirements.md, design.md, tasks.md, and spec.json so the official documentation stays in sync with the agreed feedback.

Optional: Validation Tools

Use MCP tool: validate-design
Parameters: user-authentication

Interactive design quality review with recommendations.

Use MCP tool: validate-gap
Parameters: user-authentication

Analyze gaps between requirements and existing codebase.

Command-line Options

# Show help (bunx recommended)
bunx sdd-mcp@latest --help
# or with npx
npx sdd-mcp@latest --help

# Show version
bunx sdd-mcp@latest --version

# Enable debug mode
bunx sdd-mcp@latest --debug

Global Agent Guidance (Codex / Claude Code)

If you maintain global instruction files for Codex or Claude Code agents (for example ~/.codex/AGENTS.md), include the following guidance so the assistants reliably choose sdd-mcp tools:

Map any request about "spec", "impl", or "sdd" to the matching MCP tool instead of free-form code generation.
Spell out the canonical workflow: spec-init → spec-requirements → spec-design → spec-tasks → spec-impl, and remind the agent to run spec-status before advancing phases.
Call out that steering documents (steering, steering-custom) should be refreshed whenever context feels stale, and that validate tools (validate-design, validate-gap) close the loop before shipping.
Provide ready-to-copy command examples such as Use MCP tool: spec-impl {"feature_name":"<feature-name>"} so the agent knows the exact invocation syntax.
Encourage reviewing the .kiro/ directory for outputs and approvals after each phase to keep humans in control of the process.

Development

Install Dependencies

bun install

Development Mode (with hot reload)

bun run dev

Build

bun run build

Test

# Run tests
bun test

# Run tests in watch mode
bun run test:watch

# Run tests with coverage
bun run test:coverage

Code Quality

# Lint
bun run lint

# Lint and fix
bun run lint:fix

# Format code
bun run format

OS Compatibility

OS	Support
macOS	Supported
Linux	Supported
WSL2 (Windows)	Supported
Native Windows	Not supported

Template Migration

To convert positional argument placeholders ($ARGUMENTS, $1, $2) to named placeholders ({{project_description}}, {{feature_name}}) in template files:

bun run migrate:templates

This script will:

Convert positional arguments to named placeholders
Add version field to frontmatter (if not specified)
Process all 10 template files in batch

Acknowledgments

This project is based on cc-sdd by Gota Lab. The command templates (commands/*.md) are adapted from cc-sdd under the MIT License.

License

MIT - See LICENSE for details

规格驱动开发