ScenarioSim MCP

A transparent, 100% deterministic Model Context Protocol (MCP) server that gives LLM agents a reliable what-if / scenario simulation engine.

Agents are good at describing a plan but unreliable at projecting it: they drift on multi-period arithmetic, mishandle compounding, and can't show their work. ScenarioSim offloads the simulation to an exact, explainable engine. You provide assumptions (growth rates, churn, pricing, costs, starting metrics, a time horizon); it returns projected outcomes over time, key metrics, the exact assumptions used, plus sensitivity analysis and break-even solving — each with a plain-language explanation.

Every number flows through decimal.js at 40-digit precision (never floats), so identical inputs always produce byte-identical output. The server is stateless — no database, no sessions, no clocks or randomness in the result.

This is the third product in a suite built to the same engineering standard as PrecisionCalc MCP (deterministic high-precision finance/business math) and DecisionMatrix MCP (transparent multi-criteria decision analysis): identical project structure, output philosophy, and Cloudflare Pages deployment.

🌐 Live hosted server (free, no install)

A public remote MCP server runs on Cloudflare's edge — point any Streamable-HTTP MCP client at it:

https://scenariosim-mcp.pages.dev/mcp

{ "mcpServers": { "scenariosim": {
    "type": "http", "url": "https://scenariosim-mcp.pages.dev/mcp" } } }

It runs in open mode on the free tier (no key, 20 calls/day per IP). Paid plans (Starter $12/mo · 5,000/day, Pro $39/mo · 50,000/day) are available via Stripe Checkout — buy a plan, get an API key instantly, and send it as X-API-Key. Self-host for unlimited calls with no keys. Landing page + pricing: https://scenariosim-mcp.pages.dev.

What it does

Six tools, all returning a uniform, agent-parseable envelope:

Scenario templates

Every template accepts horizon (number of periods, 1–1200) and period_label (day/week/month/quarter/year, which also sets annualization). Inputs you don't provide fall back to documented defaults; unknown inputs are ignored and reported in notes. Call list_templates for the full input/output catalog.

Consistent response envelope

Every successful response contains: status, scenario, period_label, horizon, key_results (+ key_results_detail with units and full-precision value_exact), projections, assumptions_used, methodology, notes, and a natural-language explanation.

{
  "status": "success",
  "scenario": "saas_growth",
  "period_label": "month",
  "horizon": 12,
  "key_results": {
    "ending_customers": 449.7, "ending_mrr": 26982.1, "ending_arr": 323785.2,
    "total_churned_customers": 82.4, "cumulative_revenue": 232104.6
  },
  "projections": [
    { "period": 0, "customers": 200, "mrr": 12000, "new_customers": 0, "churned_customers": 0 },
    { "period": 1, "customers": 234, "mrr": 14040, "new_customers": 40, "churned_customers": 6 }
  ],
  "assumptions_used": {
    "template": "saas_growth", "starting_customers": "200", "new_customers_per_period": "40",
    "acquisition_growth_rate": "0", "churn_rate": "0.03", "arpu": "60",
    "horizon": 12, "period_label": "month"
  },
  "methodology": {
    "model": "SaaS Growth",
    "primary_output": "ending_mrr",
    "precision": "decimal.js (40 significant digits)",
    "deterministic": true,
    "period_convention": "Period 0 is the starting state; periods 1..12 are projected. 12 month(s) per year."
  },
  "notes": ["Churn is applied to the prior period's base before new customers are added."],
  "explanation": "Starting from 200 customers and adding 40 per month (churn 3%), after 12 months you reach ..."
}

Errors never cross the tool boundary as exceptions — they come back as a structured, actionable envelope:

{
  "status": "error",
  "error": {
    "type": "unknown_template",
    "message": "Unknown scenario template 'saaas'.",
    "hint": "Available templates: saas_growth, pricing_change, churn_impact, cost_reduction, hiring_plan, cash_runway, unit_economics, marketing_funnel, compound_growth. Call list_templates for details ..."
  }
}

Design note — exact numbers: headline numbers in key_results are deterministically rounded (6 dp) for easy consumption; key_results_detail[].value_exact and assumptions_used carry full-precision strings so no precision is lost in JSON. All internal math is exact 40-digit decimal.

Project structure

scenariosim-mcp/
├── worker-src/
│   ├── index.mjs        # Cloudflare Pages Function (_worker.js): MCP over Streamable HTTP + billing routes
│   ├── engine.mjs       # The deterministic simulation engine: 9 templates + 6 tools + solver + validation
│   └── billing.mjs      # Stripe Checkout + KV-backed API keys, quota metering, webhook
├── server.mjs           # Local stdio MCP server (same engine, no network/state)
├── site/
│   ├── index.html       # Static landing / pricing / docs page
│   ├── mcp.json         # Machine-readable connection manifest
│   ├── llms.txt         # LLM-friendly summary
│   └── _worker.js       # Built bundle (esbuild output; git-ignored)
├── tests/
│   └── engine.test.mjs  # 29 core simulation-logic tests (node --test)
├── examples/
│   └── agent_example.mjs # End-to-end MCP client demo over HTTP
├── package.json         # build / deploy / dev / test scripts
├── wrangler.toml        # Cloudflare Pages config
├── .env.example         # Optional auth/rate-limit env reference
├── LICENSE              # MIT
└── README.md

Separation of concerns: engine.mjs is pure and transport-agnostic (import it directly in tests or any Node/Deno/edge runtime); index.mjs only handles the MCP JSON-RPC wiring, HTTP, CORS, and the auth/metering seam; server.mjs re-uses the same engine over stdio.

Requirements

• Node 18+ (for the build, tests, and local dev). Only two dev/runtime deps: decimal.js (math) and esbuild (bundler).
• A Cloudflare account (free tier is fine) to deploy the hosted version.

Run it locally

git clone <your-fork> scenariosim-mcp && cd scenariosim-mcp
npm install

# Run the test suite (no server needed)
npm test

# Serve the MCP endpoint locally via Wrangler (builds + runs Pages dev)
npm run dev          # -> http://127.0.0.1:8788/mcp

# Try the end-to-end client demo (hosted by default, or pass a local URL)
node examples/agent_example.mjs
node examples/agent_example.mjs http://127.0.0.1:8788

# Or run the dependency-light stdio server directly
node server.mjs

Quick manual call:

curl -s http://127.0.0.1:8788/mcp \
  -H 'content-type: application/json' \
  -H 'accept: application/json, text/event-stream' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{
        "name":"list_templates","arguments":{}}}'

Install via npm (stdio, no hosting)

Run the server locally over stdio with a single command — nothing to deploy:

npx -y scenariosim-mcp

Claude Desktop / any stdio MCP client (claude_desktop_config.json):

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

This is the same deterministic engine as the hosted server, running on your machine.

Client configuration

Cursor — ~/.cursor/mcp.json

{ "mcpServers": { "scenariosim": {
    "url": "https://scenariosim-mcp.pages.dev/mcp" } } }

Claude Desktop — claude_desktop_config.json

Claude Desktop launches stdio servers, so bridge to the HTTP endpoint with mcp-remote:

{ "mcpServers": { "scenariosim": {
    "command": "npx", "args": ["-y", "mcp-remote", "https://scenariosim-mcp.pages.dev/mcp"] } } }

VS Code — .vscode/mcp.json

{ "servers": { "scenariosim": {
    "type": "http", "url": "https://scenariosim-mcp.pages.dev/mcp" } } }

Windsurf — ~/.codeium/windsurf/mcp_config.json

{ "mcpServers": { "scenariosim": {
    "serverUrl": "https://scenariosim-mcp.pages.dev/mcp" } } }

Any Streamable-HTTP MCP client

Point it at https://scenariosim-mcp.pages.dev/mcp (or your self-hosted URL). If you enable auth, add X-API-Key (or Authorization: Bearer <key>) in the client's headers.

Tools & parameters

run_scenario(template?, inputs?, metrics?, horizon?, period_label?)

• template — one of the template ids above (aliases like saas, pricing, runway, ltv, funnel also resolve). Omit it (or pass "custom") to run a free-form model.
• inputs — the assumptions object for the template, e.g. { "churn_rate": 0.03, "arpu": 60 }. Also accepted as assumptions, or spread at the top level. Missing keys use documented defaults.
• metrics — (custom mode) array of { name, start, growth_rate?, mode? } where mode is "compound" (default, x·(1+r)ⁿ) or "linear" (x·(1+r·n)).
• horizon — number of periods to project (1–1200). Default per template (usually 12).
• period_label — day/week/month/quarter/year (default month).

sensitivity_analysis(template, variable|variables, target_metric?, variation?, steps?, values?, min?, max?, inputs?, horizon?)

Sweeps each listed input across a range (default ±variation=0.2 around the baseline, steps=5) while all others stay at baseline, recomputing target_metric (defaults to the template's primary output) at each point. Returns per-variable sweep rows, an elasticity_estimate, the output_range, and a most_influential ranking. You can also give explicit values: [...] or a min/max grid instead of variation.

break_even(template, solve_for, target_metric?, target_value, bounds?, inputs?, horizon?)

Solves for the value of solve_for (an input name) that makes target_metric equal target_value, via deterministic bisection with automatic bracket expansion. Returns required_input, change_from_baseline, achieved_metric, and residual. Assumes the metric is monotonic in the solved input over the search range; if the target can't be bracketed it returns a clean no_solution error with the achievable range. Pass explicit bounds: [lo, hi] to constrain (or fix) the search.

compare_scenarios(scenarios, compare_metric?, goal?, horizon?, include_projections?)

Runs 2–3 scenarios ({ name?, template, inputs }, or { name?, metrics } for custom) and aligns their key_results, differencing each against the first (baseline). Pass compare_metric + goal (max default | min) to rank and pick a winner. Set a shared horizon at the top level, or per-scenario.

list_templates() / health_check()

Discovery + status. No parameters.

Example tool-call payloads

Project 12 months of SaaS growth:

{ "name": "run_scenario", "arguments": {
  "template": "saas_growth",
  "inputs": { "starting_customers": 200, "new_customers_per_period": 40,
              "acquisition_growth_rate": 0.05, "churn_rate": 0.03, "arpu": 60 },
  "horizon": 12, "period_label": "month"
} }

Which lever moves ending MRR the most?

{ "name": "sensitivity_analysis", "arguments": {
  "template": "saas_growth",
  "inputs": { "starting_customers": 200, "new_customers_per_period": 40, "churn_rate": 0.03, "arpu": 60 },
  "variables": [ { "name": "churn_rate", "variation": 0.5 },
                 { "name": "arpu", "variation": 0.3 },
                 { "name": "new_customers_per_period", "variation": 0.5 } ],
  "target_metric": "ending_mrr", "horizon": 12
} }

What churn keeps 90% of customers after a year?

{ "name": "break_even", "arguments": {
  "template": "churn_impact",
  "inputs": { "starting_customers": 1000, "arpu": 60, "new_customers_per_period": 0 },
  "solve_for": "churn_rate", "target_metric": "retention_pct",
  "target_value": 0.9, "horizon": 12
} }

→ required_input ≈ 0.008742 (about 0.87%/month).

Compare growth strategies:

{ "name": "compare_scenarios", "arguments": {
  "scenarios": [
    { "name": "Base",           "template": "saas_growth", "inputs": { "churn_rate": 0.04, "new_customers_per_period": 30 } },
    { "name": "Aggressive",     "template": "saas_growth", "inputs": { "churn_rate": 0.04, "new_customers_per_period": 60 } },
    { "name": "RetentionFocus", "template": "saas_growth", "inputs": { "churn_rate": 0.015, "new_customers_per_period": 30 } }
  ],
  "compare_metric": "ending_mrr", "goal": "max", "horizon": 12
} }

Free-form (custom) model:

{ "name": "run_scenario", "arguments": {
  "metrics": [
    { "name": "revenue", "start": 10000, "growth_rate": 0.08, "mode": "compound" },
    { "name": "headcount", "start": 12, "growth_rate": 0.05, "mode": "linear" }
  ],
  "horizon": 12
} }

Deploy on Cloudflare Pages

Same pattern as PrecisionCalc / DecisionMatrix — one build step bundles worker-src/ into site/_worker.js (Pages "advanced mode" Function), then Wrangler deploys the site/ directory.

npm install
npx wrangler login          # once

# Build + deploy in one shot
npm run deploy              # esbuild -> site/_worker.js, then wrangler pages deploy

Or wire it to Git: create a Pages project, set the build command to npm run build and the output directory to site. Every push deploys automatically. The compatibility_date and project name live in wrangler.toml.

To run fully free / private, you need no bindings, secrets, or env vars — the simulation engine is stateless and the server fails open (free tier, quota disabled).

Enabling billing (optional)

Replicate these for a paid deployment:

1. KV namespace for API keys + daily usage counters, bound as SCENARIOSIM_KV in wrangler.toml (wrangler kv namespace create SCENARIOSIM_KV).
2. Stripe products/prices (subscription) — put the price IDs in [vars] (PRICE_STARTER, PRICE_PRO) and the daily limits (FREE_DAILY, STARTER_DAILY, PRO_DAILY).
3. Stripe secrets (never in the repo):
   bash

   wrangler pages secret put STRIPE_SECRET_KEY     --project-name scenariosim-mcp
   wrangler pages secret put STRIPE_WEBHOOK_SECRET  --project-name scenariosim-mcp
   

4. Webhook → create a Stripe webhook endpoint at https://<your-domain>/webhook for customer.subscription.updated + customer.subscription.deleted.

Routes wired up: /checkout?plan=starter|pro → Stripe Checkout, /success provisions and shows the API key (idempotent), /portal opens the Stripe billing portal, /webhook handles subscription lifecycle (revoke/restore), /metrics reports usage.

Auth & rate limiting

The hosted server enforces tiered quotas in worker-src/billing.mjs:

• Identity — identify() reads X-API-Key / Authorization: Bearer, looks the key up in KV, and falls back to per-IP free tier. (To add JWT/mTLS/per-org keys, change extractKey + identify only — the engine and transport are untouched.)
• Quota — consumeQuota() is a KV daily counter (resets 00:00 UTC); the single gating point in handleRpc where method === "tools/call". (Swap for a sliding-window / token-bucket in a Durable Object or Redis for per-minute limits — see the NOTE (rate limiting) comment.)
• Paywall response — over-quota / invalid / revoked keys get a structured upsell envelope with pricing + checkout URLs (agents can read and act on it).
• Usage metering — in-memory counters at /metrics.

ScenarioSim has no paid-only tools — every tool works on every tier; paid plans only raise the daily quota. To make a tool paid-only, add its name to PAID_ONLY_TOOLS in index.mjs. Because the engine is pure and stateless, none of this touches the simulation logic.

Design decisions & assumptions

• Deterministic by construction. 40-digit decimal math, ROUND_HALF_UP everywhere, period-by-period iteration (not float**n), and no clocks/randomness in results.
• Period 0 is the starting state; periods 1..horizon are projected. period_label sets the annualization factor (month → 12/yr, etc.), which is used for ARR/payroll.
• Assumptions are echoed back in full (assumptions_used) with defaults filled in, so a caller always knows exactly what was simulated.
• Counts stay fractional for precision (e.g. 233.6 customers); round to integers in your presentation layer if needed. This is stated in notes.
• Elasticity/growth models are intentionally simple and transparent (constant elasticity, constant per-period rates). They're honest first-order estimates, not econometric forecasts — the methodology block says so.
• break_even uses bisection with automatic bracket expansion and a fixed iteration budget → deterministic. It assumes monotonicity of the metric in the solved input over the range; non-monotonic/ratio metrics (with poles) return a clean no_solution rather than a wrong root. sensitivity_analysis/break_even operate on named templates (not the free-form custom model) and say so if misused.
• Errors are data, not exceptions — every tool returns status:"error" with a machine type and an actionable hint. Validation covers unknown templates/inputs/metrics, non-numeric values, bad horizons/period labels, unreachable targets, and more.
• Stateless & side-effect-free — trivially cacheable, horizontally scalable, and safe to run anywhere (Cloudflare, Node, Deno, Bun).

Testing

npm test          # node --test tests/*.test.mjs  (29 tests, no network)

The suite pins hand-verifiable arithmetic (compound growth, LTV/CAC, elasticity, runway), checks determinism, the multiple assumption-input shapes, period-label annualization, custom free-form models, the sensitivity sweep + influence ranking, the break-even solver (including the unreachable-target path), scenario comparison with goal=min, and every error path.

Roadmap (post-MVP)

1. More templates: LBO/DCF, inventory & cash-conversion cycle, ad-spend ROAS, cohort retention.
2. Monte-Carlo mode: distributions on inputs → confidence bands on outcomes (seeded, still deterministic).
3. Multi-variable (grid) sensitivity and tornado charts alongside one-at-a-time.
4. Break-even on the free-form custom model and on multiple simultaneous inputs.
5. Per-key usage dashboard + Durable-Object quotas for stronger consistency.

License

MIT — see LICENSE.