This update makes NeqSim easier to use as a process-simulation and fluid-behaviour tool from agentic MCP clients. The main change is not a new thermodynamic model. It is a stronger contract between agents and NeqSim: agents can now discover what is available, fetch schemas and examples, build valid inputs, understand units, inspect validation and benchmark maturity, and route results through review gates.

The external-simulator interoperability item is intentionally not included in this update. CAPE-OPEN, FMI, and commercial simulator bridge work remains a separate future task.

What Improved

Area Before After this update
Response parsing Tools returned useful JSON, but response shape varied by runner High-use calculation, automation, and lifecycle responses expose a standard MCP envelope with apiVersion, tool, data, provenance, validation, qualityGate, and warnings
Process execution Agents could call process simulations directly and see runtime failures runProcess can run validation first and return validation context with the process result
Tool discovery Capability metadata focused on the highest-use runners getCapabilities describes all 63 MCP tools, including detailed descriptors for high-use tools and generic descriptors for the rest
Schemas Detailed schemas covered the high-use tool set Every advertised MCP tool has an input and output schema URI; high-use tools keep detailed schemas, remaining tools get generic contract schemas
Examples Core examples covered common flash, process, validation, and safety patterns Every advertised MCP tool can now return at least a contract-level starter example through the tool category
Setup guidance Agents had to infer workflow setup from prose getCapabilities and neqsim://setup-templates expose 26 setup templates across flash, process, PVT, flow assurance, lifecycle, safety, reporting, and uncertainty workflows
Units and JSON fields Process JSON and units were documented but not fully machine-readable The capability manifest now includes processJsonContract, unitSystem, and equipmentPropertyOntology sections
Trust and review Benchmark maturity existed but was not tied into a larger agent policy The manifest now includes benchmarkRegistry, modelLifecycle, optimizationUncertaintyWorkflows, and safetyGatePolicy metadata
Regression coverage Tests checked selected schemas and capability descriptors Tests assert that every advertised tool has schemas, examples, validation metadata, response-contract metadata, and capability coverage

New Capability Sections

getCapabilities now includes these machine-readable sections in addition to the existing domain, equipment, trust, and workflow metadata:

Section Why it matters for agents
toolCapabilities Maps MCP tool names to schema names, runner classes, categories, required fields, units, examples, templates, limitations, validation, and response-contract metadata
processJsonContract Lists legal process root fields, fluid fields, equipment fields, connection fields, supported equipment types, stream ports, common properties, and unit encoding rules
capabilityGraph Provides graph nodes and edges linking domains, tools, schemas, examples, and governance contracts
equipmentPropertyOntology Helps agents assemble flowsheets by exposing equipment ports, common settable properties, dynamic support, and automation variables
benchmarkRegistry Tells agents where to fetch maturity, validation cases, accuracy bounds, reference data, and known limitations before relying on results
unitSystem Gives preferred unit dimensions and encoding rules so generated payloads avoid silent unit mistakes
automaticFlowsheetBuilder Describes a staged workflow for turning a process description into validated runProcess JSON
optimizationUncertaintyWorkflows Describes supported sensitivity, model-risk, scenario, and P10/P50/P90 patterns
modelLifecycle Describes inspect, mutate, snapshot, restore, diff, and audit fields for stateful model workflows
safetyGatePolicy Keeps safety, relief, LOPA, SIL, risk, flare, HAZOP, and barrier outputs behind explicit engineering review gates

Practical Agent Flow

An agent can now follow a reliable workflow without relying on hidden knowledge:

  1. Call getCapabilities.
  2. Pick the relevant tool and setup template.
  3. Fetch neqsim://schemas/{tool}/input and neqsim://examples/tool/{tool}.
  4. Build explicit unit-bearing input JSON.
  5. Call validateInput where applicable.
  6. Execute the selected tool.
  7. Parse data, provenance, validation, qualityGate, and warnings.
  8. Fetch benchmark trust or safety-gate metadata before using results in a design claim.

This is especially important for process simulation and fluid behaviour because model choice, units, component naming, operating envelope, and validation maturity all change how trustworthy a result is.

Validation Added

The focused tests now cover the discovery contract rather than only a few happy paths:

Together, these tests guard against a common agentic failure mode: advertising a capability that has no schema, no example, no validation story, or no predictable response shape.

What This Does Not Claim

This update completes the MCP foundation for agentic NeqSim workflows: advanced workflows are now discoverable, schema-backed, unit-aware, benchmark-aware, lifecycle-aware, and review-gated. Some higher-level workflows, such as P&ID-to-flowsheet generation and full Monte Carlo or Pareto optimization, are exposed as standardized agent workflows rather than certified autonomous execution engines.

Engineering-critical results still need professional review, benchmark checks, standards context, and independent verification where the task requires it. The remaining work is deeper execution automation and broader industrial validation, not basic MCP capability discovery.