Dev Team MCP — Autonomous AI Software Development Team

Submit a ticket → get production-ready code.

An autonomous MCP server that runs a full software development pipeline: Planner → Coder → Tester → Reviewer → Delivery

Runs on port 8002 alongside:

• DevOps MCP (:8000) — Docker, K8s, AWS, Git
• Dev Agent MCP (:8001) — Code generation in 18 languages

---

Architecture

┌─────────────────────────────────────────────────────────────────────┐
│  Claude Code / Copilot Chat                                         │
│                                                                     │
│  MCP Stdio Adapters:                                                │
│  ├── devops   → mcp-server      :8000  (DevOps — infra/git)        │
│  ├── dev      → mcp-dev-agent   :8001  (Code gen — 18 languages)   │
│  └── devteam  → dev-team-mcp    :8002  (Dev Team — full pipeline)  │
│                      │                                              │
│            ┌─────────▼─────────┐                                   │
│            │   ORCHESTRATOR    │                                    │
│            └─────────┬─────────┘                                   │
│                      │                                              │
│    ┌─────────────────┼─────────────────┐                           │
│    ▼                 ▼                 ▼                 ▼          │
│ PLANNER           CODER            TESTER           REVIEWER        │
│ Decompose         Generate         Write &          Security +      │
│ ticket into       production       run tests        perf + style    │
│ plan + stack      code files       fix failures     checklist       │
│    │                 │                 │                 │          │
│    └─────────────────┴─────────────────┴─────────────────┘         │
│                                │                                    │
│                      ┌─────────▼─────────┐                         │
│                      │  DELIVERY.md      │                         │
│                      │  git repo + tag   │                         │
│                      └───────────────────┘                         │
└─────────────────────────────────────────────────────────────────────┘

Each agent uses:

• LLM (Ollama / OpenAI / Anthropic) for reasoning
• Dev Agent :8001 for high-quality code generation and reviews
• DevOps MCP :8000 for git operations and deployments

---

Quick Start

1. Install

cd ~/dev-team-mcp
make install

2. Configure

cp .env.example .env
# Edit .env — choose LLM_PROVIDER (default: ollama)

3. Run

# Start this server
make dev        # Dev server on :8002 with auto-reload

# Or start all three MCP servers at once
make start-all

4. Verify

curl http://localhost:8002/health | jq

---

How to Use

Submit a ticket (async)

curl -X POST http://localhost:8002/ticket \
  -H "Content-Type: application/json" \
  -d '{
    "title": "Add JWT authentication to FastAPI app",
    "description": "Add JWT-based auth with login endpoint, token refresh, and protected routes. Use python-jose.",
    "language": "python",
    "framework": "fastapi",
    "priority": "high",
    "labels": ["feature", "security"]
  }'

Response:

{
  "ticket_id": "a1b2c3d4e5f6",
  "status": "pending",
  "message": "Ticket accepted. Pipeline started: Planner -> Coder -> Tester -> Reviewer. Poll GET /ticket/a1b2c3d4e5f6 for status."
}

Poll for status

curl http://localhost:8002/ticket/a1b2c3d4e5f6 | jq

{
  "ticket_id": "a1b2c3d4e5f6",
  "status": "done",
  "output_path": "/home/user/dev-team-mcp/workspace/add-jwt-auth-a1b2c3d4e5f6",
  "review_score": 88,
  "plan_summary": "Implement JWT authentication..."
}

Get all generated files

curl http://localhost:8002/ticket/a1b2c3d4e5f6/artifacts | jq

Get a specific file

curl http://localhost:8002/ticket/a1b2c3d4e5f6/artifacts/auth/jwt.py | jq .content

Blocking / sync mode (for small tasks)

curl -X POST http://localhost:8002/ticket/sync \
  -H "Content-Type: application/json" \
  -d '{"title": "Hello world script", "description": "Python script that prints hello world and current time"}'

---

MCP Tools (for Claude Code)

Once registered in ~/.claude/settings.json, you can use these tools directly in Claude:

Tool	Description
devteam_submit_ticket	Submit a ticket for autonomous development
devteam_submit_sync	Submit and wait (blocking)
devteam_ticket_status	Check pipeline progress
devteam_ticket_logs	Stream agent log entries
devteam_list_tickets	List all tickets
devteam_list_artifacts	List generated files
devteam_get_artifact	Get a specific file content
devteam_review_report	Get code review score + checklist
devteam_cancel_ticket	Cancel a ticket
devteam_health	Health check

---

Pipeline Stages

1. Planner Agent

• Reads the ticket and produces a structured JSON plan
• Chooses the minimal viable tech stack
• Breaks work into numbered steps with file targets
• Flags security concerns (OWASP Top-10)
• Defines test strategy and CI/CD approach

2. Coder Agent

• Generates production-ready code for every file in the plan
• Uses Dev Agent :8001 when available (falls back to direct LLM)
• Applies OWASP mitigations (input validation, parameterised queries, no hardcoded secrets)
• Auto-retries with alternative approach on failure

3. Tester Agent

• Generates unit + integration tests for every source file
• Runs tests via the language's native test runner
• On failure: asks LLM to fix the implementation and re-runs (loop)
• Tracks coverage percentage

4. Reviewer Agent

• Full security audit (uses Dev Agent :8001 + LLM bundle review)
• Performance analysis
• Documentation completeness check
• Production readiness checklist (12 criteria)
• Auto-fixes medium/low issues
• Approves (score ≥ 75) or rejects with detailed feedback

Delivery

• All files written to workspace/<slug>-<ticket_id>/
• Git repository initialised with commit history
• DELIVERY.md — full handover document
• production-ready-<id> git tag applied when approved

---

Output Structure

workspace/add-jwt-auth-a1b2c3d4e5f6/
├── DELIVERY.md          ← Human-readable handover
├── auth/
│   ├── jwt.py           ← JWT implementation
│   ├── routes.py        ← Protected routes
│   └── test_jwt.py      ← Unit tests
├── main.py              ← FastAPI app
├── requirements.txt
└── .github/
    └── workflows/
        └── ci.yml       ← GitHub Actions CI

---

Configuration

Env var	Default	Description
LLM_PROVIDER	ollama	ollama / openai / anthropic
OLLAMA_MODEL	mistral	Any Ollama model (codestral, llama3.1, etc.)
OPENAI_API_KEY	—	Required when LLM_PROVIDER=openai
ANTHROPIC_API_KEY	—	Required when LLM_PROVIDER=anthropic
DEVOPS_MCP_URL	http://localhost:8000	DevOps MCP server URL
DEV_AGENT_URL	http://localhost:8001	Dev Agent MCP server URL
DEV_TEAM_PORT	8002	This server's port
MAX_RETRIES	3	Max LLM retries per step
MAX_TEST_FAILURES	5	Max test fix iterations
WORKSPACE_DIR	./workspace	Where generated repos land

---

Running Tests

make test

---

Full 3-Server Stack

# Terminal 1: DevOps MCP (infrastructure)
cd ~/mcp && make dev

# Terminal 2: Dev Agent MCP (code gen)
cd ~/mcp-dev-agent && make dev

# Terminal 3: Dev Team MCP (autonomous pipeline)
cd ~/dev-team-mcp && make dev

Or use the convenience target:

cd ~/dev-team-mcp && make start-all

---

Generated by Dev Team MCP — autonomous AI software development pipeline