⚡ OptimEngine

Operations Intelligence for AI Agents — L1 → L3 in one conversation.

11 MCP tools that optimize, quantify risk, forecast, and prescribe. From production scheduling to Monte Carlo simulation. From delivery routing to Pareto frontiers. Ask in natural language, get optimal decisions.

---

🤖 Use with Claude (60 seconds)

No installation. No code. Works on claude.ai (Free, Pro, Max, Team, Enterprise).

1. Open claude.ai
2. Click + → Integrations
3. Add custom integration
4. Paste this URL:

https://optim-engine-production.up.railway.app/mcp/sse

Claude discovers all 11 tools automatically. Try asking:

> "Schedule 5 production orders on 4 machines, minimize delays"
>
> "Optimize delivery routes for 6 clients with 2 trucks"
>
> "What happens if dosing time increases 30%? Run sensitivity analysis"
>
> "I have historical data for the last 8 weeks. Forecast next month and recommend actions"

---

🧠 What Is OptimEngine?

An Operations Intelligence Engine — not a wrapper, not a chatbot. A computational decision brain that solves NP-hard optimization problems and quantifies risk. Powered by Google OR-Tools CP-SAT and Routing solvers.

4 Intelligence Levels

| Level | Capability | Question It Answers |
|-------|-----------|-------------------|
| L1 Deterministic | Scheduling, Routing, Packing, Validation | What's the optimal plan? |
| L2 Uncertainty | Sensitivity, Robust, Stochastic | How fragile is this plan? |
| L2.5 Multi-Objective | Pareto Frontier | What's the best trade-off? |
| L3 Prescriptive | Forecast → Optimize → Risk → Advise | What should I do and why? |

11 MCP Tools

L1 — Deterministic Optimization
- optimize_schedule — Flexible Job Shop (FJSP) with precedence, setup times, priorities, 4 objectives
- optimize_routing — CVRPTW with capacity, time windows, distance matrix, drop visits
- optimize_packing — Multi-dimensional bin packing with weight, volume, groups
- validate_schedule — Find overlaps, precedence violations, eligibility errors

L2 — Optimization under Uncertainty
- analyze_sensitivity — Parametric perturbation, elasticity, risk ranking
- optimize_robust — Worst-case / percentile protection, price of robustness
- optimize_stochastic — Monte Carlo + CVaR with 4 distributions

L2.5 — Multi-Objective
- optimize_pareto — 2-4 competing objectives, trade-off analysis, correlation

L3 — Prescriptive Intelligence
- prescriptive_advise — 4 forecast methods, 3 risk appetites, confidence intervals, action items

Infrastructure
- health_check — System status
- root — Server info, capabilities, tool listing

---

📊 Live Demo Results

Every number below comes from a real call to OptimEngine. Zero mock data.

Digital Twin Decisionale — NovaCosm (Cosmetics Manufacturer)

Full production-to-delivery chain: 6 lines, 8 orders, 5 brand clients.

| Phase | Tool | Result |
|-------|------|--------|
| Plant diagnosis | optimize_schedule | 575 min makespan, 2 late, Line 2 bottleneck (82.6%) |
| Cycle time forecast | prescriptive_advise | 3 rising trends (+1.1-1.65%/week), +5.6% makespan in 4 weeks |
| What-if: cross-line | optimize_schedule | Move 1 product to Line 1 → tardiness -65%, lines balanced |
| Risk profile | optimize_stochastic | 50 Monte Carlo, CV 4.5%, 100% feasible |
| Client doubles orders | optimize_schedule | Without investment: 530 min tardiness, 50% late |
| + New line investment | optimize_schedule | With Line 2B: 0 tardiness, 478 min, 8/8 on-time |
| Manual schedule check | validate_schedule | 4 violations found (overlaps + precedence) |
| Risk ranking | analyze_sensitivity | Serum dosing most critical (score 11.3, elasticity 0.227) |
| Delivery routing | optimize_routing | 6 clients, 2 trucks, 70 km, all time windows met |
| Truck loading | optimize_packing | 8/8 pallets, 0 excluded, route-constrained |

Strategic decisions generated: cross-line authorization (saves 125 min/day), maintenance alert (Line 2 degrading), investment quantification (Line 2B enables client growth).

BevDistri (F&B HoReCa Distribution)

| Tool | Result |
|------|--------|
| optimize_routing | 18 clients, 2/3 vehicles used, 132 km, all windows met |
| prescriptive_advise | +21% Modena demand in 4 weeks, decision deadline identified |
| optimize_packing | 13 items, 2 bins, 0 drops, 97% utilization alert |
| analyze_sensitivity | Hotel demand can double without route split |

---

🔧 For Developers

MCP Configuration (Claude Desktop, Cursor)

{
  "mcpServers": {
    "optim-engine": {
      "command": "npx",
      "args": [
        "supergateway",
        "--sse",
        "https://optim-engine-production.up.railway.app/mcp/sse"
      ]
    }
  }
}

Direct API

curl -X POST https://optim-engine-production.up.railway.app/optimize_schedule \
  -H "Content-Type: application/json" \
  -d '{
    "jobs": [
      {
        "job_id": "ORD-001",
        "priority": 8,
        "due_date": 480,
        "tasks": [
          {"task_id": "dosing", "duration": 90, "eligible_machines": ["line_A", "line_B"], "setup_time": 15},
          {"task_id": "filling", "duration": 60, "eligible_machines": ["line_A", "line_D"], "setup_time": 10}
        ]
      }
    ],
    "machines": [{"machine_id": "line_A"}, {"machine_id": "line_B"}, {"machine_id": "line_D"}],
    "objective": "minimize_makespan"
  }'

Orchestration Pattern: Routing → Packing

When combining routing and packing (e.g., delivery logistics), use this pattern:

1. Call optimize_routing → get routes with vehicle-to-client assignments
2. Partition items by route → each vehicle's items based on routing output
3. Call optimize_packing per vehicle → separate packing per truck/van

This ensures pallet assignments match delivery routes. See the NovaCosm demo for a complete example.

---

🌐 Available On

| Platform | Link |
|----------|------|
| Claude.ai | Add as custom integration (instructions above) |
| MCPize | mcpize.com/mcp/optim-engine |
| Apify Store | apify.com/hearty_indentation/optim-engine |
| LobeHub | lobehub.com/mcp/michelecampi-optim-engine |
| mcp.so | mcp.so/server/optim-engine |
| Railway | optim-engine-production.up.railway.app |
| ERC-8004 | Agent #22518 on Base L2 |
| Landing Page | optim-engine-landing.vercel.app |

---

📈 Numbers

| Metric | Value |
|--------|-------|
| Solver modules | 9 |
| MCP tools | 11 |
| Tests passing | 121 |
| Intelligence levels | 4 (L1, L2, L2.5, L3) |
| Forecast methods | 4 |
| Stochastic distributions | 4 |
| Scheduling objectives | 4 |
| Routing objectives | 4 |
| Risk appetites | 3 |
| ERC-8004 Agent | #22518 (Base L2) |
| Capital invested | €0 |

---

🏗️ Architecture

                    ┌─────────────────────────────────┐
                    │         Claude / AI Agent        │
                    │    (natural language interface)   │
                    └──────────────┬──────────────────┘
                                   │ MCP Protocol
                    ┌──────────────▼──────────────────┐
                    │      OptimEngine v8.0.0          │
                    │      FastAPI + MCP Server        │
                    ├─────────────────────────────────┤
                    │  L1 Deterministic                │
                    │  ├─ Scheduling (CP-SAT FJSP)     │
                    │  ├─ Routing (OR-Tools CVRPTW)    │
                    │  ├─ Packing (CP-SAT)             │
                    │  └─ Validator                    │
                    ├─────────────────────────────────┤
                    │  L2 Uncertainty                  │
                    │  ├─ Sensitivity Analysis          │
                    │  ├─ Robust Optimization           │
                    │  └─ Stochastic (Monte Carlo)     │
                    ├─────────────────────────────────┤
                    │  L2.5 Multi-Objective             │
                    │  └─ Pareto Frontier               │
                    ├─────────────────────────────────┤
                    │  L3 Prescriptive Intelligence     │
                    │  └─ Forecast → Optimize → Advise  │
                    └─────────────────────────────────┘
                         Google OR-Tools · Python
                         Railway · ERC-8004 Base L2

---

🤝 Consulting & Custom Integration

Need OptimEngine configured for your specific operations? Production scheduling, logistics optimization, risk analysis for your plant?

I build Digital Twin Decisional systems — from scheduling diagnosis to strategic what-if analysis. The solver runs in seconds; the domain expertise makes it useful.

Michele Campi — Operations Intelligence Engineer
- GitHub: @MicheleCampi
- 7+ years operations controlling in cosmetics contract manufacturing
- Built OptimEngine solo: 11 tools, 4 intelligence levels, 121 tests, zero capital

---

License

MIT — use it freely. The code is open; the intelligence design is the moat.