Spaces:
Running
title: EPANET MCP Server
emoji: 💧
colorFrom: blue
colorTo: indigo
sdk: docker
app_port: 7860
pinned: false
license: mit
short_description: EPANET/ePyT water-network modelling over MCP (HTTP + SSE)
💧 EPANET MCP Server
A remote Model Context Protocol server that exposes
EPANET water-distribution modelling — powered by ePyT —
to any MCP-capable client. Load .inp models, run hydraulic/quality simulations, modify
parameters, and generate what-if scenarios in natural language.
It wraps the upstream epanet-mcp-server
(44 tools) and serves it over both modern transports at once, behind an optional API key.
Endpoints
| Transport | Path | Clients |
|---|---|---|
| Streamable HTTP | /mcp |
HuggingChat, Codex, Perplexity, Gemini, ChatGPT (dev mode) |
| HTTP + SSE | /sse |
Claude Desktop (via mcp-remote), other legacy clients |
| Health probe | /health |
uptime checks (no auth) |
Base URL once deployed: https://<owner>-<space-name>.hf.space
Authentication
Set a Space secret named CLIENT_API_KEY. When present, every request to /mcp,
/sse, and /messages/ must send it:
Authorization: Bearer <YOUR_KEY>
(X-API-Key: <YOUR_KEY> is also accepted.) Leave the secret unset to run open — useful for
ChatGPT connectors, which only support OAuth or no-auth. / and /health are always public.
Networks
ePyT ships 52 benchmark networks (Net1/2/3, Hanoi, Anytown, L-TOWN, Balerma, …) — call
list_bundled_networks to see them. Custom networks are bundled under
/home/user/app/networks/ and loaded by absolute path, e.g.:
load_network("/home/user/app/networks/net3.inp", "net3")
Bundled here: net3.inp, net3_PDD.inp, ky4.inp, ky10.inp, Net6.inp, BIWS.inp.
Smoke test
smoke_test.py runs the full MCP handshake (initialize → tools/list → tools/call) over
both transports and checks the auth gate:
python smoke_test.py --base-url https://<owner>-<space-name>.hf.space --api-key <YOUR_KEY>
See CLIENTS.md for copy-paste connection configs for each client.
Built on epanet-mcp-server by Applied-AI-Eurecat (MIT).