Actuarial chain-ladder reserving — built for Claude.

A Model Context Protocol server that hands Claude eight tools for loss-triangle math, Mack stochastic estimation, and assumption diagnostics. Hook it up in one minute, then ask in plain English.

Install  → Pro tier · $29

Try it now

uvx mcp-chainladder

What it does

Eight tools, picked automatically — you don't have to know their names.

Chain-ladder projection

Volume- and simple-averaged factors, CDF to ultimate, latest-diagonal extraction, IBNR per AY + total — the whole machinery in one call.

Mack stochastic estimation

Distribution-free standard errors and coefficients of variation per accident row and total, computed exactly per Mack (1993).

Assumption diagnostics

Calendar-year sign test, Spearman independence, inflation slope, and standardised-residual outlier flagging — all the assumption checks Mack (1994) recommends.

CSV-aware

Hand Claude a path. It parses your triangle (NA, dashes, thousand-comma quirks all handled) and runs the chain ladder on it.

Cumulative ↔ incremental

Switch triangle shapes losslessly. Round-trips preserve unobserved cells exactly so you don't lose information mid-analysis.

Pure local, zero network

No HTTP calls, no analytics, no telemetry. Every triangle stays on your machine. Pop-up the source and check — it's under 1,200 LOC.

Free vs Pro

Free covers the core projection + stochastic + diagnostics. Pro adds interpretive analytics and additional reserving methods.

Tool Free Pro
compute_chain_ladder — full projection
project_triangle — fill lower-right
mack_stochastic — Mack (1993) SE/CV
mack_diagnostics — raw Mack tests
parse_csv_triangle — load from CSV
to_incremental / to_cumulative
sample_triangle — textbook demo
interpret_diagnostics — verdict labels + plain-English summaries
sensitivity_analysis — drop-one-out IBNR impact ranking
tail_extrapolation — exponential + inverse-power fits
bornhuetter_ferguson — BF with CL side-by-side
compare_methods — CL + BF in one call

What a conversation looks like

Once installed, ask Claude in plain English. The MCP makes the right tool calls under the hood.

You
I have a paid-loss triangle at ~/Desktop/motor.csv. Run the chain
ladder, flag any diagnostic concerns, and recommend whether I can
publish this as a reserve.

Claude (calls parse_csv_triangle + compute_chain_ladder + interpret_diagnostics)

Loaded a 10×10 cumulative triangle. Volume-weighted factors give:
  • Paid to date:    49,458 USD
  • Ultimate:        65,883 USD
  • IBNR:            16,425 USD
  • Mack SE total:   ±354.61 USD (CV 0.54%)

Diagnostics:
  • Calendar-year (Tarbell):       Z=0.00, p=1.000 — no evidence
  • Independence (Spearman):       Z=-0.35, p=0.726 — no evidence
  • Inflation slope:               β=0.053, p=1.3e-12 — STRONG evidence
  • Outliers (|residual| > 2):     none

Recommendation: link ratios drift upward by ~5%/AY (statistically
strong). Before publishing, weight only the last 3–5 AYs and compare
against a Bornhuetter-Ferguson cross-check.

Pricing

Free for everyone. Pro is a one-time purchase that unlocks BF, sensitivity, tail fits, and interpretive diagnostics.

Chain Ladder Pro

$29 perpetual
  • Bornhuetter-Ferguson + CL side-by-side
  • Interpretive diagnostics — plain-English verdicts
  • Drop-one-out sensitivity ranking
  • Tail extrapolation (exp / inv-power)
  • All future Pro tools, free of charge
  • Email support, 30-day refund
Buy Pro  →