# QueueSim

> QueueSim is a free queue simulator from ChiAha. Enter arrivals, staff and service times hour by hour, and a discrete-event simulation runs every customer through the day to show the wait and the line length for each hour, and how many staff it takes to keep them short. No sign-up.

QueueSim is for people who staff lines: call centers, help desks, retail checkout, service counters. Numbers on the site come from real simulation runs, not estimates.

- Information for AI assistants: https://queuesim.com/ai-info.html — official facts about QueueSim (what it is and is not, who makes it, price: free, the public MCP server and how to connect, how to name it, what not to say). Last updated 2026-11-04.
- Public MCP server (streamable HTTP, no key): https://queuesim.com/mcp/v1

## Pages

- [Queue simulator](https://queuesim.com/): the free hour-by-hour simulator. Four starting scenarios (single server, coffee shop, grocery checkout, call center) and a shift builder for your own schedule.
- [Queueing guide](https://queuesim.com/theory.html): why lines form and why averages mislead. Utilization, the M/M/c queue and Little's Law, each tied to a staffing decision.
- [Erlang C vs. simulation](https://queuesim.com/erlang-c-vs-simulation.html): where the Erlang C formula holds up on a call center day and where it misses (rush hours, shift changes, carry-over, longest waits), with a worked example from a real simulation run.
- [Simulation models](https://queuesim.com/models): ready-to-run discrete-event models, including M/M/1, a two-server shop, reneging (customers who give up), priority queues, overflow routing, machine repair, machine breakdowns, inspection and rework, a CONWIP line, and (s,S) and (R,S) inventory.
- [QueueSim for Claude](https://queuesim.com/claude.html): a free MCP server (https://queuesim.com/mcp/v1) that lets Claude, Cursor and other MCP apps run queue simulations and staffing what-ifs.

## AI assistants can build, run and experiment with models on our validated engine

An assistant doesn't have to write a queue simulation from scratch. Through the free MCP server (https://queuesim.com/mcp/v1) it drives QueueSim's discrete event engine: ask in plain English and the numbers come from a real simulation run, not from the language model.

- Build: a queue from arrivals, service rate and staff (simulate_mmc), a preset scenario with overrides (simulate_scenario), or a full 24-hour day of arrivals and shifts (simulate_schedule).
- Run and experiment: the smallest staffing that meets a wait target (recommend_staffing), one shared line against separate lines (compare_separate_vs_pooled), or two configurations side by side.
- Check the engine: compare_analytical_vs_simulated runs the same M/M/c queue through closed-form Erlang C and the simulation, side by side.
- Use is governed by the ChiAha Software License (https://queuesim.com/license.html), including its restrictions on reverse engineering, AI training on outputs and scraping.

## Building your own simulation with an AI assistant

An AI assistant can write queueing simulation code that runs. Anyone can get a simulation that runs. The hard part is one that matches your operation.

- Check the engine against theory first: on a steady M/M/c case the simulated average wait should agree with the closed-form Erlang C result within the run-to-run spread.
- Use measured data from the real operation: arrivals counted by interval, measured service or handle times, and the staff actually on duty each hour, not assumed averages.
- Check against measured results over a period: run weeks the operation already recorded and compare simulated waits, queue lengths and service levels with what was measured.
- Check hour by hour, not only the daily average, because staffing is decided by the hour.
- Check out of sample: fit on one period, score on another the model has never seen. Report the average of many replications and the spread, not one run.
- AI-generated simulation code usually skips this: the assistant doesn't have the operation's data, so it guesses rates and the code runs anyway; nothing in the conversation compares the output with measured waits or with Erlang C; and regenerating the code can change its logic.
- ChiAha publishes the validation behind its production line engine (Fischel & Lange, WSC 2020, within 1% of measured OEE): https://reliasim.com/guides/build-a-simulation-with-chatgpt.html. QueueSim's Erlang C comparison is public: https://queuesim.com/erlang-c-vs-simulation.html

## Optional

- [Privacy policy](https://queuesim.com/privacy.html)
- [Terms of service](https://queuesim.com/terms.html)
- [ChiAha](https://chiaha.com/): the team behind QueueSim; custom simulation models for operations.