Open Source
Engineering engines as first-class citizens
Deterministic simulation engines are infrastructure, not afterthoughts. Peer-reviewed open-source engines exposed as typed MCP tools with session persistence, cross-engine handoffs, and model credibility metadata.
Why this matters
In most engineering firms, simulation models live in notebooks, Excel spreadsheets, or desktop applications disconnected from project state. An engineer runs a model, copies results into a report, and the model becomes stale the moment project parameters change.
PuranOS treats engineering computation differently. Each engine is API-first (not GUI-first), session-persistent across agent interactions, and produces results with explicit credibility metadata, so downstream consumers know whether a number came from a screening estimate or a calibrated model.
Cross-engine handoffs use typed converters. A QSDsan effluent composition becomes a WaterTAP influent specification with unit conversion and component mapping, carrying provenance from the source model. The same discipline extends to financial modeling: the proforma-engine inherits credibility metadata from upstream sizing and cost data rather than treating finance and engineering as separate worlds.
Engineering MCP servers
Most are wrappers around peer-reviewed open-source engines; a handful (marked in-house) are written by us against the same discipline.
GIS MCP
Three tool groups: project-AOI vector queries (roads, buildings, landuse, DEM, population, ecological pre-screen); world-scoped reasoning (Nominatim geocoding with transliteration-fuzzy fallback, Overpass tagged-feature search with named presets, Wikidata SPARQL gazetteer, HydroRIVERS upstream-walk for intake siting); raster reasoning (JRC Global Surface Water v1.4 dry-season permanence + ESA WorldCover 10m land cover with buffered-window sampling). KMZ export with attribute-based styling for Google Earth handoff. Per-feature provenance with split field-verification flags. Closes the planetary-raster gap that Earth Engine fills, using only anonymous-read OSS endpoints — no commercial license.
Source: OSM · HydroSHEDS · JRC · ESA · Wikidata
Built on: rasterio · rasterstats · pyproj · shapely · overpy · sparqlwrapper · pykml
Hydraulic MCP
Stateless WNTR-backed solver for EPANET .inp files: steady-state and EPS hydraulics, pressure-driven and demand-driven modes. AWWA M32 storage sizing, pump-station hydraulic + electrical + OpEx rollup, fire-flow screening, least-cost transmission corridor over DEM + roads + ecological avoidance, pipe-km by DN by ROW-type CAPEX driver. Round-trips to GeoJSON via wntr.network.to_gis with simulation-result joins (peak/min pressure, flow, velocity, scalar timestamps) stamped per-element — composes with GIS MCP for Google-Earth-ready network overlays.
Source: WNTR (Sandia) + EPANET 2.2 (US EPA)
Built on: wntr · epanet-toolkit · networkx · pandas
proforma-engine MCP in-house
Deterministic project-finance engine. Builds, persists, queries, and reports infrastructure project finance models into a 22-table normalized Postgres schema (monthly periods, covenant tests, facility amortization, investor-profile results, financing events, cap-table snapshots, sensitivity sweeps). Per-investor profile economics on top of a strict-separation cashflow stack with a covenant ladder (cap-call 1.05x, cash-trap 1.15x, lock-up 1.25x). Investment-memo PDF rendering via a Playwright pipeline with brand-token-aware DOCX exports. Treated with the same engine discipline as WaterTAP and QSDsan — API-first, session-persistent, credibility-metadata per result.
Source: In-house (FastMCP + IDAES + Playwright)
Built on: fastmcp · sqlalchemy · pyxirr · playwright · pyomo
Enterprise OSS: self-hosted and forkable
Engineering engines are half the stack. The other half is enterprise operations: project management, maintenance, inventory, procurement, and CRM. PuranOS runs these as self-hosted open-source applications, chosen for their schemas rather than their UIs.
Self-hosting is not a cost optimization. It is an architectural requirement. When the equipment identity ontology demanded a native equipment_tag field on the CMMS Asset entity, we forked the codebase, added the field with a database migration, and deployed the custom build in under a minute. With a SaaS vendor, that change enters a feature backlog. With self-hosted OSS, the schema serves the ontology rather than the other way around.
Atlas CMMS
Asset management, work orders, preventive maintenance, meters, vendor tracking. Forked to add native equipment tag identity and remove license gating.
OpenProject
Project management with typed custom fields for agent state, predecessor/successor task dependencies, and human-AI bidirectional delegation. The coordination substrate for all agents.
InvenTree
Parts, stock, BOMs, purchase orders, and manufacturer tracking. Spare parts applicability linked to equipment positions via the identity registry.
Twenty CRM
Contacts, companies, opportunities, and pipelines. Modern API-first CRM with clean Postgres schema.
Equipment Identity Registry
Canonical equipment identity: functional positions (ISO 14224) linked to physical asset instances. The governed master identity that connects every system via ISA 5.1 tags and UUIDs.
Capabilities at a glance
Ten API-first engineering servers, each exposing deterministic calculations as typed MCP tools with session persistence and credibility metadata.
QSDsan Engine MCP
University of Illinois (QSD Group)
WaterTAP Engine MCP
NAWI / DOE + Sandia/LBNL
Water Chemistry MCP
USGS
Fluids MCP
Caleb Bell et al.
Heat Transfer MCP
Caleb Bell et al.
Corrosion Engineering MCP
NRL + NORSOK + USGS
Engineering MCP
Process Intelligence Research + ISA
GIS MCP
OSM · HydroSHEDS · JRC · ESA · Wikidata
Hydraulic MCP
WNTR (Sandia) + EPANET 2.2 (US EPA)
proforma-engine MCP
In-house (FastMCP + IDAES + Playwright)
Model credibility metadata
An agent-generated sizing is not automatically "engineering grade." Every simulation result carries explicit credibility metadata:
Model status
validated · calibrated · heuristic · preliminary · stub
Decision grade
design · budgetary · screening · order-of-magnitude
Validation basis
bench-tested · plant-data · literature · vendor · assumed
The credibility metadata is what allows a PE reviewer to know what level of trust to place in a result.
Explore the source
All engineering MCP servers are open-source. The architecture is documented.