DEV Community

Cover image for Claude Code + Crevo: Achieving Efficient & High-Quality SDD (Spec-Driven Vibe-Coding)
Aurakl Crevo
Aurakl Crevo

Posted on

Claude Code + Crevo: Achieving Efficient & High-Quality SDD (Spec-Driven Vibe-Coding)

#architecture #softwaredevelopment #mcp #programming

I. How to Configure Claude Code with Our Crevo MCP Tool

Objective: To enable Claude Code to access Crevo via MCP (for reading/writing the 7 key documents exported by Crevo, triggering Crevo's generation/update interfaces, etc.).

Working Configuration

Recommended (CLI) Method

# Add Crevo's MCP service to Claude Code
claude mcp add --transport http aurakl-crevo https://crevo-mcp.aurakl.ai/mcp?key=your-api-key
Enter fullscreen mode Exit fullscreen mode
  • aurakl-crevo is the name for the MCP server being added (which can be referenced in Claude Code).
  • Replace your-api-key with the API Key obtained from your Crevo console.

Or, Create **.mcp.json**** in the Project Root**
Place .mcp.json in the project root (for project-level scope) or in the global config (for global scope):

{
  "mcpServers": {
    "aurakl-crevo": {
      "serverUrl": "https://crevo-mcp.aurakl.ai/mcp?key=your-api-key"
    }
  }
}
Enter fullscreen mode Exit fullscreen mode

Verification

claude mcp list
Enter fullscreen mode Exit fullscreen mode

II. How to Configure Crevo's Two Slash Commands: /crevo and /update

Create the /crevo command

mkdir -p ~/.claude/commands
echo "Use Crevo to create the complete seven key documents (User Stories, PRD, System Architecture Design, Business Process Design, Database Design, API Design, Development Plan)." > ~/.claude/commands/crevo.md
Enter fullscreen mode Exit fullscreen mode

After executing, simply type in the Claude Code chat box:

/crevo User Management System
Enter fullscreen mode Exit fullscreen mode

Claude Code will automatically call the Crevo MCP service to generate the complete seven documents.

Create the /update command

echo "Automatically update the Crevo document system (including PRD, Architecture, API, Test Cases, etc.) based on the latest development results." > ~/.claude/commands/update.md
Enter fullscreen mode Exit fullscreen mode

Usage:

/update User Management System v1.2
Enter fullscreen mode Exit fullscreen mode

Crevo will automatically sync the design documents based on Claude Code's development execution, achieving an intelligent, closed-loop update of the specifications.

III. How to Use Crevo to Create Documents

Crevo is responsible for turning product goals into "structured specifications". Below are copy-pasteable example snippets (real, exportable Markdown / OpenAPI / SQL) that can be generated by Crevo for use in Claude Code.

Document Overview (using a "User Management System" as an example)

.spec/
├── user_story.md
├── prd.md
├── architecture.md
├── process_design.md
├── database.md
├── api_spec.md
└── dev_plan.md
Enter fullscreen mode Exit fullscreen mode
  1. User Stories
  2. PRD (Product Requirements Document)
  3. System Architecture Design
  4. Business Process Design (Process / Flow)
  5. Database (DB) Model (Schema / Migration)
  6. API (OpenAPI Spec)
  7. Development Plan (dev_plan)

Example 3.1 — User Stories **docs/user_story.md**

# User Stories

## US-001: User Registration
As a: Unregistered user
I want: To register an account by submitting my email and password
So that: I can log in and use the platform's features

Acceptance Criteria:
- Frontend provides an email/password registration form
- Email format validation, password must be at least 8 characters
- Send an activation email
Enter fullscreen mode Exit fullscreen mode

Example 3.2 — PRD **docs/prd.md**

# PRD: User Management Service

Objective:
- Provide account registration, login, user profile management, and role-based access control (RBAC).

Scope:
- Email/Password registration, OAuth login
- User Profile CRUD
- Role & permission management
- Audit logging and security monitoring

Non-Goals:
- SSO integration (can be considered in a future iteration)
Enter fullscreen mode Exit fullscreen mode

Example 3.3 — System Architecture **docs/architecture.md**** (Excerpt)**

Components:
- API Gateway
- Auth Service (JWT, OAuth)
- User Service (profile, preferences)
- Role Service (RBAC)
- DB: PostgreSQL + Redis (cache)
- CI/CD: GitHub Actions / Jenkins
Enter fullscreen mode Exit fullscreen mode

Example 3.4 — Business Process **docs/process_design.md**** (Mermaid)**

sequenceDiagram
  User->>Frontend: submit(email, password)
  Frontend->>API: POST /api/v1/register
  API->>AuthService: validate, hash password
  AuthService->>DB: insert user
  AuthService->>EmailService: send activation email
  User->>EmailService: click activation link
  EmailService->>DB: set is_active = true
Enter fullscreen mode Exit fullscreen mode

Example 3.5 — DB **docs/database.sql**

CREATE TABLE users (
  id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
  email VARCHAR(255) NOT NULL UNIQUE,
  password_hash TEXT NOT NULL,
  created_at TIMESTAMP WITH TIME ZONE DEFAULT now(),
  is_active BOOLEAN DEFAULT false
);

CREATE TABLE roles (
  id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
  name VARCHAR(100) NOT NULL UNIQUE
);

CREATE TABLE user_roles (
  user_id UUID REFERENCES users(id),
  role_id UUID REFERENCES roles(id),
  PRIMARY KEY (user_id, role_id)
);
Enter fullscreen mode Exit fullscreen mode

Example 3.6 — API **docs/api_spec.yaml**** (OpenAPI Snippet)**

openapi: 3.0.3
info:
  title: User Management API
  version: "1.0.0"
paths:
  /api/v1/register:
    post:
      summary: Register user
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/RegisterRequest'
      responses:
        '200':
          description: registration accepted
components:
  schemas:
    RegisterRequest:
      type: object
      required:
        - email
        - password
      properties:
        email: { type: string, format: email }
        password: { type: string, minLength: 8 }
Enter fullscreen mode Exit fullscreen mode

Example 3.7 — Dev Plan **docs/dev_plan.md**

# Dev Plan: user-management

iterations:
- id: iter-01
  goal: basic registration & login
  tasks:
    - t-101: implement users table and migrations
    - t-102: implement POST /api/v1/register
    - t-103: create unit tests for register flow

- id: iter-02
  goal: RBAC & profile management
  tasks:
    - t-201: implement roles table and user_roles mapping
    - t-202: implement role assignment API
    - t-203: implement user profile CRUD
Enter fullscreen mode Exit fullscreen mode

IV. Creating Claude Code SubAgents Based on Development Needs

The Development Plan (Dev Plan) generated by Crevo will be parsed by Claude Code into a series of development tasks.
Each task corresponds to a SubAgent, which is defined as a Markdown file in the .claude/agents/ directory.

SubAgents are not automatically generated from the documents, but are created based on the task type (development, testing, code review, audit, scheduling, etc.).
This approach allows Claude Code to execute tasks by assigning roles.

SubAgent Example Collection

dev-agent.md

# Development Agent
Responsible for completing specific code development based on the design documents generated by Crevo.

## Input
- prd.md
- api_spec.md
- db_design.md

## Behavior
1. Read the tasks for the current iteration
2. Generate or update the corresponding code files
3. Output the code files and commit log
Enter fullscreen mode Exit fullscreen mode

test-agent.md

# Test Agent
Automatically generate and execute test cases based on the API design and user stories.

## Input
- api_spec.md
- user_story.md

## Behavior
1. Generate Jest test files
2. Run tests and record results
3. Output test report to `reports/`
Enter fullscreen mode Exit fullscreen mode

review-agent.md

# Code Review Agent
Review the latest code commits for security, style consistency, and potential logic issues.

## Input
- Latest commit log
- Security guidelines document

## Behavior
1. Perform static analysis and security review
2. Output a review report with remediation suggestions
Enter fullscreen mode Exit fullscreen mode

audit-agent.md

# Audit Agent
Monitor system changes, dependency updates, and sensitive configurations.

## Behavior
- Check dependency changelogs
- Output compliance report
Enter fullscreen mode Exit fullscreen mode

scheduler-agent.md

# Scheduler Agent
Responsible for scheduling the execution order and dependencies of SubAgents.

## Behavior
1. Read `dev_plan.md`
2. Execute independent tasks concurrently
3. Aggregate all results and report back to Crevo
Enter fullscreen mode Exit fullscreen mode

This way, Claude Code's MCP layer can dynamically assign tasks to different SubAgents based on Crevo's development plan and context, achieving modular execution.

V. Executing Iterative Development and Generating Code

Once the seven documents and SubAgents are ready, simply execute:

#claude
Please help me complete the first iteration as planned in dev_plan.md
Enter fullscreen mode Exit fullscreen mode

Claude Code will:

  1. Read the Iteration 1 tasks from dev_plan.md;
  2. Plan development tasks by referencing the iteration plan and the relevant content within the 7 key documents;
  3. Call the scheduler-agent to schedule;
  4. Assign tasks to dev-agent, test-agent, etc.;
  5. Automatically generate code and run tests;

VI. Crevo's Specification-Driven Vibe Coding

Crevo's core philosophy is that all development tasks must originate from a 'Specification' (Spec).
The Claude Code and MCP architecture turn these specs into the contextual source of truth, and all of the AI's actions take place within this 'semantic field (Vibe)'.

Feature Comparison:

Traditional AI Coding SDD Vibe Coding
One-shot generation, lacks contextual consistency Continuous collaboration based on complete specs and context
Manual debugging and manual fixes Automatic iteration, testing, and document synchronization
Code and documentation are separate Documentation as the semantic source for code
No feedback loop Complete design-implement-validate-update cycle

VII. Advantages of SDD Vibe Coding

  • Efficient Alignment: Product, architecture, development, and testing are all aligned on the same structured document, reducing meetings and repetitive clarifications.
  • Reproducible and Auditable: Every action by a SubAgent (generation, commit, test) has logs, PRs, and CI records.
  • Automated Closed Loop: Forms a continuous optimization cycle, from requirements to deployment and back to monitoring data flowing into the docs.
  • Reduced Human Error: The mapping from spec to implementation is deterministic: OpenAPI → code skeleton; User Stories → test cases.
  • Suitable for Scaled Teams: Ideal for large projects with multi-team collaboration that require clear interfaces and contracts.
  • Supports Chinese/English Users: Crevo's multi-lingual output + Claude Code's universal agent design allow for barrier-free collaboration for international teams.

Further Reading

Top comments (0)