Overview

The simulator now ships with an HTTP MCP (Model Context Protocol) server, exposing the automata engine to AI assistants and other programmatic clients. Through a single JSON RPC 2.0 endpoint at POST /api/mcp, a model can simulate machines, convert between formalisms, and generate shareable canvas links without ever touching the UI.

What You Can Do

Simulate Automata

Six tools run input strings against the core machine types and report whether the input is accepted, along with the path taken:

simulate_fa — DFA / NFA
simulate_pda — pushdown automata
simulate_ndpa — nondeterministic pushdown automata
simulate_tm — Turing machines (single- and multi-tape)
simulate_mealy — Mealy machines
simulate_moore — Moore machines

Convert Between Formalisms

Three tools bridge the classic equivalences:

regex_to_nfa — compile a regular expression into an NFA
nfa_to_dfa — determinize an NFA via subset construction
fa_to_regex — extract a regular expression from a finite automaton

Inspect Machines

get_transition_table — the full transition table for an automaton
get_automata_info — states, alphabet, and structural metadata

Generate Shareable Canvas Links

Six build_*_link tools (build_fa_link, build_pda_link, build_ndpa_link, build_tm_link, build_mealy_link, build_moore_link) encode a complete workspace state into a URL fragment. Opening the link loads the machine straight onto the canvas, ready to edit and run.

How It Works

Everything is served over a standard JSON-RPC 2.0 interface, so any MCP-compatible client can list the available tools and call them by name. The generated links use the canvas #load= hash convention: a mount hook decodes the base64 payload, reconstructs the automaton, and renders it on load no server round trip required.

Why It Matters

The MCP server turns the simulator into a tool an assistant can use, not just describe. Ask a model to build an automaton for a language, and it can validate the construction, convert it, and hand you a link to the finished machine on the canvas.