An autonomous software-delivery platform where AI agents build, validate, and ship entire production applications — governed by humans, audited end-to-end, and isolated by design.
Conceived, architected, and built by Abdel Alahdal
A product of Infragate Solutions LTD — United Kingdom
Engineering teams spend 80 % of their time on repetitive scaffolding, code review, testing, and deployment — not on the 20 % that actually differentiates their product. Existing "AI coding tools" only solve the typing problem. They don't build, they don't validate, and they certainly don't ship.
Give AI agents a structured state machine, a sandboxed environment, a validation pipeline, and a human escalation path — and they can deliver entire production applications on their own.
MPAOP is the first platform to treat AI-driven development as a supervised, auditable workflow rather than an ad-hoc chat session. Every stage transitions through a formal state machine. Every artifact is stored and signed. Every decision — whether taken by Claude, GPT, or a human reviewer — is captured in an immutable audit log.
| Surface | URL | Purpose |
|---|---|---|
| Dashboard | streamtvlive.cloud | Human operator console — Replit-inspired workspace |
| REST API | api.streamtvlive.cloud/v1 | 11 route groups, Zod-validated, bearer-token auth |
| MCP Server | mcp.streamtvlive.cloud/mcp | Streamable HTTP transport — 8 tools |
| Per-Project Apps | *.streamtvlive.cloud |
Dynamic subdomain routing via Traefik + Let's Encrypt |
Every endpoint returns 401 Unauthorized without a valid project bearer token (Argon2id-hashed, constant-time verified).
flowchart TB
subgraph Human["Human Operators"]
U[Product Owner]
end
subgraph Edge["Edge Layer"]
T[Traefik v3<br/>+ Let's Encrypt]
end
subgraph Control["Control Plane"]
W[Next.js 14<br/>Dashboard]
A[Fastify 4 API<br/>REST + MCP]
O[Autopilot Orchestrator<br/>Architect ↔ Executor Loop]
end
subgraph AI["AI Providers"]
C[Claude Code CLI<br/>via @anthropic-ai/sdk]
G[GPT Reviewer<br/>via OpenAI SDK]
end
subgraph Pipeline["Validation Pipeline"]
V1[TypeScript Validator]
V2[ESLint Validator]
V3[Vitest Runner]
V4[QA Agent<br/>Playwright]
end
subgraph Data["Data Plane"]
DB[(PostgreSQL 16<br/>Prisma ORM)]
R[(Redis 7<br/>BullMQ Queues)]
M[(MinIO<br/>S3-compatible)]
end
subgraph Projects["Per-Project Workspaces"]
P1[Project A<br/>Docker --internal]
P2[Project B<br/>Docker --internal]
P3[Project N<br/>Docker --internal]
end
U --> T
T --> W
T --> A
W --> A
A --> O
A --> DB
A --> R
A --> M
O --> C
O --> G
O --> Pipeline
Pipeline --> P1
Pipeline --> P2
Pipeline --> P3
C --> P1
C --> P2
C --> P3
Every unit of AI work moves through an explicit finite-state machine with optimistic concurrency and atomic claim semantics:
stateDiagram-v2
[*] --> QUEUED
QUEUED --> RUNNING_CLAUDE: claimed by executor
RUNNING_CLAUDE --> STORING_RESULT: Claude completes
STORING_RESULT --> GPT_REVIEWING: artifacts uploaded
GPT_REVIEWING --> APPROVED: score ≥ threshold
GPT_REVIEWING --> REVISION_NEEDED: actionable fixes
GPT_REVIEWING --> ESCALATED: ambiguous / blocked
REVISION_NEEDED --> RUNNING_CLAUDE: retry (new attempt)
ESCALATED --> WAITING_HUMAN_APPROVAL: queued for review
WAITING_HUMAN_APPROVAL --> HUMAN_APPROVED
WAITING_HUMAN_APPROVAL --> HUMAN_REJECTED
APPROVED --> DONE
HUMAN_APPROVED --> DONE
HUMAN_REJECTED --> REVISION_NEEDED
DONE --> [*]
| Capability | Why It Matters |
|---|---|
| Autopilot: Architect ↔ Executor loop | Two specialised AI roles. The Architect plans tasks, the Executor implements them, the Reviewer validates. No prompt engineering required. |
| Multi-Project Isolation | Every project lives in its own Docker --internal network with dedicated Postgres, MinIO bucket, and Redis key-prefix. Zero cross-contamination. |
| Dual Protocol Surface | The exact same 8 tools are exposed over both MCP Streamable HTTP (for Claude Code) and a mirrored REST API (for any HTTP client). |
| Immutable Audit Log | Every action — from stage.claimed to project.approved — is appended to the AuditLog table with actor, timestamp, and JSON payload. No backfilling. |
| Token-Scoped Access | Per-project bearer tokens are Argon2id-hashed at rest and verified in constant time. No shared global keys. |
| Production Validators | TypeScript, ESLint, Vitest, and Playwright QA agent — all run in isolated workers via BullMQ. Flaky mocks are rejected. |
| Live Preview + Subdomain Router | Every project gets a real subdomain (project-xyz.streamtvlive.cloud) serving its built app — not a status page. |
| Human Review Console | Replit-inspired workspace with 10 integrated tabs: Overview, Stages, Chat, Preview, QA, Autopilot, Files, Logs, Activity, Review. |
|
Control Plane
|
AI Layer
|
Data Plane
|
Infrastructure
|
mpaop/
├── apps/
│ ├── api/ # Fastify REST + MCP server (11 route groups)
│ ├── web/ # Next.js operator dashboard (19 pages)
│ └── workers/ # BullMQ validation + executor (concurrency-safe)
├── packages/
│ ├── core/ # Business logic — the brains
│ │ ├── autopilot/ # Architect ↔ Executor loop
│ │ ├── claude/ # Claude SDK integration + 8 tools
│ │ ├── executor/ # Stage execution engine
│ │ ├── validators/ # tsc / eslint / vitest / QA
│ │ ├── review/ # GPT reviewer integration
│ │ ├── provisioning/ # Per-project Docker orchestration
│ │ ├── deployer/ # Subdomain router + live preview
│ │ ├── qa/ # Playwright QA agent
│ │ ├── chat/ # Conversational interface
│ │ ├── tokens/ # Argon2id token management
│ │ ├── notifications/ # Activity feed + alerts
│ │ └── workflow/ # State machine + transitions
│ ├── db/ # Prisma schema (17 models)
│ ├── mcp/ # MCP Streamable HTTP transport
│ ├── providers/ # Claude + OpenAI adapters
│ └── shared/ # Cross-cutting types + utilities
├── scripts/
│ ├── verify-claude-integration.ts # 22-step integration test
│ └── verify-isolation.ts # 9-point isolation audit
├── docker-compose.yml # Postgres, Redis, MinIO, Traefik, Workers
├── ecosystem.config.cjs # PM2 production process manager
├── turbo.json # Turborepo pipeline
└── pnpm-workspace.yaml
| Metric | Value |
|---|---|
| TypeScript lines of code | 18 442 |
| TS + TSX source files | 111 |
| Prisma data models | 17 |
| API route groups | 11 |
| Core packages | 15 sub-domains |
| Git commits (curated history) | 49 |
| Dashboard pages | 19 |
| Validators | 4 (tsc · eslint · vitest · Playwright QA) |
| AI tools exposed | 8 (REST + MCP) |
Note: MPAOP is proprietary software. Running it requires a license from Infragate Solutions LTD. The instructions below are provided for evaluation and due-diligence purposes.
- Node.js ≥ 20
- pnpm ≥ 9.15
- Docker + Docker Compose
- A Claude API key (
ANTHROPIC_API_KEY) - An OpenAI API key (
OPENAI_API_KEY) — used by the reviewer
# 1. Clone and install
git clone https://github.com/alahdal262/mpaop-platform.git
cd mpaop-platform
pnpm install
# 2. Copy the env template and fill in secrets
cp .env.example .env.local
$EDITOR .env.local
# 3. Start local infrastructure (Postgres, Redis, MinIO)
docker compose up -d platform-db redis minio
# 4. Generate Prisma client and apply schema
pnpm --filter @mpaop/db generate
pnpm --filter @mpaop/db push
# 5. Run the control plane
pnpm dev # Turborepo orchestrates api + web + workers
# 6. Open the dashboard
open http://localhost:3002Production is managed via PM2 with ecosystem.config.cjs and fronted by Traefik. See the deployment brief under NDA.
Abdel Alahdal is the sole conceiver, architect, and author of MPAOP. He designed the state machine, authored every line of the control-plane TypeScript, built the 19-page Replit-inspired dashboard, deployed the production infrastructure, and shipped the platform end-to-end as a one-founder team.
MPAOP is the materialisation of a single conviction: that software delivery can be automated without sacrificing engineering rigour. Where other AI-coding tools stop at autocomplete, MPAOP orchestrates entire production releases — and does so with the auditability regulated industries require.
Abdel is the Founder of Infragate Solutions LTD, a UK-registered software firm specialising in AI-native infrastructure for enterprises that cannot afford to be wrong.
"The future isn't AI writing code. The future is AI shipping code — safely, repeatedly, and under adult supervision. MPAOP is that supervisor." — Abdel Alahdal, Founder
Infragate Solutions LTD Registered in England and Wales · United Kingdom infragatesolutions.com
Infragate Solutions LTD builds AI-native infrastructure for organisations where correctness, auditability, and isolation are non-negotiable. Our focus is on turning experimental AI capabilities into production systems that pass enterprise procurement.
We build: Orchestration platforms · Agentic workflows · MCP servers · LLM-grounded internal tools · Enterprise AI compliance layers.
For licensing, partnership, or commercial enquiries: https://infragatesolutions.com
MPAOP is licensed under a proprietary license — see LICENSE for the full text.
Copyright © 2026 Infragate Solutions LTD. All rights reserved. Conceived and authored by Abdel Alahdal.
Viewing this repository on GitHub is permitted for evaluation and due diligence. Any commercial use, redistribution, or derivative work requires a written agreement with Infragate Solutions LTD.