Siyana

صِيانة — Arabic for "maintenance."

Open-source Model Context Protocol server delivering region-tuned facilities-management asset knowledge — specs, fault resolution, and PPM schedules — directly inside any LLM client. Launching GCC-first.

Facilities teams in the Gulf answer the same questions every day: What are the specs for this asset? How do I resolve this fault? What's the right preventive-maintenance schedule for it in this climate? This server puts that knowledge — tuned for the Middle East's heat, dust, and hard water — one question away in Claude, Cursor, or any MCP-compatible client.

⚠️ Guidance, not gospel. The knowledge base is synthesised from public manufacturer information and open standards, then GCC-tuned. It is not a substitute for the OEM manual, a competent engineer's judgement, or local regulations. Always verify before acting.

🌱 A living, community-grown knowledge base

Siyana is an open ecosystem — the knowledge base grows with the FM community, not just one vendor. Anything missing? Request it in one click, no coding required:

| If you want to… | Open this issue | |------|------| | Add equipment we don't cover (e.g. a fire pump) | 🧰 New asset / equipment | | Add a brand to assets we have (e.g. "Add Trane HVAC equipment") | 🏷️ Add a brand / manufacturer | | Propose a new trade or service (fire, BMS, lifts, cleaning…) | 🧱 New trade / service | | Start a new region (UK, US, ANZ, EU…) | 🌍 New region dataset | | Flag something wrong or outdated | 🩹 Report inaccurate data |

Prefer to contribute the knowledge directly? Even better — see CONTRIBUTING.md. Every entry is paraphrased and cited (never copyrighted text), reviewed by a maintainer, and shipped in the next release. Run npm run stats to see current coverage.

What it covers (v1)

Three trades, ~16 core asset types, each with specifications, GCC-tuned PPM schedules, fault-resolution guides, and cited sources:

| Trade | Assets | |-------|--------| | HVAC | Air-cooled chiller, AHU, FAHU, VRF/VRV, split AC, cooling tower, fan coil unit | | Electrical | LV distribution board, diesel generator, UPS, distribution transformer, ACB/LV switchgear | | Plumbing | Booster pump set, submersible drainage/sewage pump, electric water heater, GRP water tank |

Not yet included: market/labour rates (planned for v2). See Roadmap.

Tools

| Tool | What it does | |------|--------------| | list_asset_categories | List the trades and every available asset type with its id. Start here. | | search_assets | Fuzzy-search by keyword, asset type, brand, or model (e.g. "Daikin VRV"). | | get_asset_spec | Full specs for an asset type, plus cited brand/model spec sheets and region context. | | get_ppm_schedule | Region-tuned preventive-maintenance tasks, frequencies, and the standard behind each. | | get_fault_resolution | Diagnose a symptom → likely causes, step-by-step resolution, safety notes. |

Live weather-aware tools (need network; disable with SIYANA_OFFLINE=1):

| Tool | What it does | |------|--------------| | get_weather | Current conditions + daily forecast + elevated dust/PM10 for a location. | | estimate_cooling_energy | Indicative cooling kWh (and optional cost / peak TR) for a space, using live degree-days × a cited region/building-type benchmark. | | get_maintenance_advisory | Turns the forecast into proactive maintenance actions (heatwave → coil clean, dust → filters), optionally tied to a specific asset's PPM tasks. |

Quick start

Requires Node.js ≥ 18. The server runs over stdio and needs no configuration or network access.

npx siyana

Claude Desktop / Claude Code

Add to your MCP config (claude_desktop_config.json, or .mcp.json for Claude Code):

{
  "mcpServers": {
    "siyana": {
      "command": "npx",
      "args": ["-y", "siyana"]
    }
  }
}

Cursor

In ~/.cursor/mcp.json (or project .cursor/mcp.json):

{
  "mcpServers": {
    "siyana": {
      "command": "npx",
      "args": ["-y", "siyana"]
    }
  }
}

Then ask, for example:

• "What's the PPM schedule for a Daikin VRV running in Dubai?"
• "My air-cooled chiller keeps tripping on high pressure in summer — how do I resolve it?"
• "How often should I clean a GRP water tank in the UAE, and why?"

Why GCC-tuned?

Generic global maintenance data under-serves the Gulf. This knowledge base bakes in:

• Extreme heat — high-ambient derating, more frequent condenser-coil cleaning, battery-life impacts.
• Dust & sand — shorter filter life, radiator/coil fouling, sand-trap louvres.
• Hard water — scaling of heaters, cooling-tower fill, and pump seals; descaling regimes.
• Local norms — references to DEWA, SASO, Dubai Municipality, and Civil Defence requirements.

Regions & scope

Siyana ships one dataset per region, and the server is scoped to a region as an environment property. Each region is a self-contained dataset — a GCC chiller and a UK chiller are different files with their own specs, PPM, and faults. Tools only ever read from the active region's dataset.

data/
  gcc/                          ← region = dataset (the only one populated today)
    equipments/<category>/
      category.md               ← frontmatter (specs · PPM · faults) + Markdown body (guidance)
      <brand>/<model>.md        ← model specs + link to the OEM's official manual
    services/    resources/    benchmarks/   ← further pillars (rolling out)
  uk/  us/  anz/  eu/           ← add a region by adding a folder; no rebrand

Siyana's knowledge is organised into four region-scoped pillars: equipments (category → brand → model, live today), services (trades like housekeeping/glass-cleaning), resources (FM roles + indicative rates), and benchmarks (by facility type).

• Scope with SIYANA_REGION (gcc | uk | us | anz | eu); defaults to gcc. The server reads from data/<region>/ and every tool is confined to it.
• Override per call: each tool also accepts an optional region argument to read another region's dataset.
• list_asset_categories reports the active region, the default, and which regions have data today.
• Asking for a region with no dataset returns a clear "no asset … in region" message — never silent or wrong.

// Scope a server instance to the UK dataset
{
  "mcpServers": {
    "siyana-uk": {
      "command": "npx",
      "args": ["-y", "siyana"],
      "env": { "SIYANA_REGION": "uk" }
    }
  }
}

Today only the gcc dataset is populated. See the roadmap.

Weather-aware intelligence

Static guidance becomes situational when it knows the local weather. Siyana integrates the free, open, no-key Open-Meteo API so it can:

• Estimate cooling energy for a space from live climate data — e.g. "how much energy will my 10,000 sqft Dubai office need this summer?" It pulls cooling-degree-days for the location and applies a cited region/building-type energy-intensity benchmark.
• Plan maintenance around the forecast — a 48 °C heatwave next week → prioritise condenser-coil cleaning; a dust event → bring filter changes forward.

estimate_cooling_energy({ location: "Dubai", area_sqft: 10000, building_type: "office", period: "summer" })
→ ~74,000 kWh (range 55k–98k) · summer CDD 2,186 · indicative ±~25%, with assumptions + sources

Indicative, not an MEP load calculation. Estimates use the degree-day × benchmark method with wide error bars. The benchmarks in reference/benchmarks/ are community-maintained — improve them via PRs.

Design: the five knowledge tools are fully offline and deterministic — only the three weather tools use the network, with timeouts, caching, and graceful failure. Run SIYANA_OFFLINE=1 to disable the weather layer entirely (e.g. air-gapped sites); the knowledge tools keep working. Weather data © Open-Meteo.com (CC BY 4.0).

How the knowledge stays up to date

• Every asset file carries sources and a last_reviewed date.
• New brand/model or revised practice → open a pull request → CI validates → maintainer review → release. npx users get it on the next run.
• npm run freshness reports entries overdue for review; a scheduled CI job opens "review needed" issues.

Found an asset we're missing, or something out of date? Open an issue or PR — see CONTRIBUTING.md.

Development

npm install
npm run dev        # run the server from source (tsx)
npm test           # run the test suite
npm run validate   # validate every data file against the schema
npm run gen-schema # regenerate schema/asset.schema.json from src/schema.ts
npm run build      # compile to dist/
npm run inspect    # build + open the MCP Inspector against the server

The data shape is defined once in src/schema.ts (zod) and mirrored to schema/asset.schema.json for contributors and CI.

Roadmap

• v2 — market & labour rates: SQFT- and trade-wise FM staff/service rate benchmarks for GCC markets (sourcing under evaluation).
• Global regions: Siyana is region-agnostic by design — GCC ships first, with UK / US / ANZ / EU profiles to follow as a region dimension of the data (no rebrand, same package).
• More trades: fire & life-safety, BMS, vertical transport.
• Arabic-language content.
• Optional HTTP/SSE transport for hosted deployments.

License

• Code (everything outside data/): Apache-2.0.
• Data (the data/ knowledge base): CC BY 4.0.

Maintained by the community.