ReliaStats MCP v1

A free MCP server from ChiAha exposing reliability-statistics theory and interpretation tools. Designed for LLM clients (Claude Desktop, Cursor, ChatGPT Apps SDK, any MCP-compatible client) and human exploration alike.

Endpoint

JSON-RPC 2.0 over Streamable HTTP. Standard MCP initialize / tools/list / tools/call methods.

Tools (11)

Theory (doc-backed, deterministic):

• explain_reliability_basics — R/F/h, MTBF/MTTF/MTTR, availability, bathtub curve, RBD basics
• explain_distributions_for_reliability — Weibull centre-stage, β-as-failure-mode
• explain_advanced_reliability_patterns — censoring, MLE, GOF, PI vs CI, ALT, Bayesian
• explain_pi_vs_ci_for_validation — the Validation scatter's red/blue/teal reference bands

Pure-math interpretive (closed-form, stateless):

• interpret_weibull_shape(beta, eta?) — failure-mode regime + (if η provided) MTTF / B-life
• weibull_summary(beta, eta) — MTTF, B10/50/90, hazard at t
• compute_availability(mtbf, mttr) — A = MTBF / (MTBF + MTTR)
• system_reliability(components, structure) — series or parallel RBD
• recommend_distribution(symptoms) — symptom → distribution shortlist

Cross-MCP bridge (soft-reference catalog — no proxying):

• list_paired_models — paired sandbox catalog (today: bottling line)
• describe_bottling_line — the bottling-line paired model: topology, tracks, build sequences, workflow

Anti-fabrication

Every numeric result is closed-form reliability math (deterministic). Every text result is a quoted section from ChiAha-authored docs (docs/reliability-theory.md, docs/paired-model-bottling-line.md). Nothing comes from LLM training-data recall.

Sandbox

The MCP is the interpretation layer. To actually fit distributions, compare two interrupt files, or validate a simulation against historian data, use the ReliaStats sandbox at reliastats.com/app.

License

Read-only public surface. Free to use. Not for redistribution. See /license.