Skip to main content
Glama
sauna-mcp-server

sauna-mcp-server

by divelux

sauna-mcp-server

A small Model Context Protocol server that gives an AI assistant four sauna tools: plan a session, size a heater, look up a bathing tradition, look up a sauna word.

It runs on your own computer over stdio. Every tool is read-only. Nothing is sent over the network, nothing is written to disk, and no state is kept between calls.

Made by DIVELUX, a small family brand in Irvine, California that makes handmade wool sauna hats.

Tool

What it does

Where the numbers come from

plan_session

Rounds, minutes per round, breaks, final cool-down and glasses of water for a sauna session

The rules of the Sauna Session Planner (sauna-planner-core)

size_heater

Heater size for a room: a kW range for an electric heater, a room volume for a wood-burning stove

Harvia and HUUM published sizing rules, as in the Sauna Heater Size Calculator

lookup_tradition

28 bathing traditions (Finnish sauna, banya, hammam, onsen, jjimjilbang, temazcal and more) and 12 kinds of bath headwear

The open dataset Sauna Traditions of the World (CC BY 4.0)

lookup_term

145 sauna and bathhouse words in 25 languages, with plain-English meanings

The open Sauna Glossary (CC BY 4.0)

Not medical advice. plan_session returns planning numbers for healthy adults: conservative, widely repeated sauna conventions, not research findings and not targets. Every plan ends with the line "Not medical advice. ... Step out if you feel dizzy, sick or faint. ... talk to your doctor first." The server does not know anyone's health.

Heater sizes are estimates. Every size_heater answer ends with "Always confirm with the heater manufacturer and a licensed electrician."

Set it up

You need Node.js 20 or newer.

Claude Code

claude mcp add sauna -- npx -y sauna-mcp-server

Add --scope user before the name to have it in every project, or --scope project to share it with your team through .mcp.json. Check it with claude mcp list, or /mcp inside Claude Code.

The same server as a .mcp.json entry:

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

Claude Desktop

Open Settings from the Claude menu, go to Developer, and click Edit Config. That opens claude_desktop_config.json:

  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

  • Windows: %APPDATA%\Claude\claude_desktop_config.json

  • Linux: ~/.config/Claude/claude_desktop_config.json

Add the server (keep any servers already listed), save, then quit Claude Desktop completely and start it again:

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

From a copy of this repository

npx -y sauna-mcp-server works once the package is on npm. Until then, or to run your own copy:

git clone https://github.com/divelux/sauna-mcp-server.git
cd sauna-mcp-server
npm install --omit=dev

Then use node with the full path to server.mjs in place of npx:

claude mcp add sauna -- node /full/path/to/sauna-mcp-server/server.mjs
{ "mcpServers": { "sauna": { "command": "node", "args": ["/full/path/to/sauna-mcp-server/server.mjs"] } } }

Any other MCP client that starts local stdio servers works the same way: the command is npx -y sauna-mcp-server, with no arguments and no environment variables.

Related MCP server: finops-framework-mcp

Things to ask

  • "Plan a sauna session for me: Finnish sauna at 175 °F, I go a few times a month, and I have 90 minutes."

  • "My sauna room is 6 by 5 feet with a 7 foot ceiling and a glass door. What size electric heater do I need?"

  • "What is the difference between a Russian banya and a Finnish sauna?"

  • "What does löyly mean? And what is the Russian word for the sauna whisk?"

The tools

plan_session

Argument

Values

Default

type

dry, loyly, banya, steam, infrared

dry

temp

thermometer reading in unit; readings outside the usual range for the type are moved to the nearest end

a typical temperature for the type

unit

f, c

f

level

new, regular, seasoned

new, the most cautious

total_minutes

15 to 180

60

include_ladder

true to add the heat ladder, an illustration of air temperature from ceiling to floor

false

Called with {"type": "dry", "temp": 175, "unit": "f", "level": "regular", "total_minutes": 90} it answers:

Sauna session plan: Finnish, dry at 175 °F, regular bather, 90 minutes available.

0:00  Round 1, 11 min
0:11  Cool-down, 11 min
0:22  Round 2, 11 min
0:33  Cool-down, 11 min
0:44  Round 3, 11 min
0:55  Final cool-down, 10 min

In the heat: 3 rounds of 11 min (33 min in total).
Breaks: 11 min between rounds, then a 10 min final cool-down.
Time used: 65 of 90 min (25 to spare). Spare time is not turned into more heat.
Bench: either bench.
Water: 6 glasses of 8 fl oz (48 fl oz, 1440 ml): 1 before, 2 in the breaks, 3 after.

Not medical advice. These are planning numbers for healthy adults, not targets. Step out if you feel dizzy, sick or faint. If you have a health condition or are not sure the heat is right for you, talk to your doctor first.

The structured result holds plan (exactly what plan() of sauna-planner-core returns), adjustments (what the planner changed or assumed, in words), share_url (the same plan in the web planner), disclaimer, and ladder when asked for.

The rules, in short: experience sets a band of minutes per round; a hotter room means a shorter round; nobody gets more than 20 minutes at a stretch in a hot room; breaks are at least as long as the rounds; the plan never runs over the time given and never turns spare time into more heat. They are written out in full, with their one outside source (Harvard Health Publishing), in the sauna-planner-core README.

size_heater

Argument

Values

Default

unit

ft (decimal feet and square feet) or m (meters and square meters)

ft

length, width, height

inside size of the hot room; in feet use decimals, 6 ft 6 in is 6.5

required

glass_area

uninsulated glass: door, window, glass wall

0

masonry_area

uninsulated stone, tile, brick or concrete

0

log_walls

true if the inside walls are bare log

false

heater

electric or wood

electric

Called with {"length": 6, "width": 5, "height": 7, "glass_area": 12} it answers:

Electric heater: 5.1 to 7.3 kW for a calculated volume of 257 cu ft (7.3 m³).
High end: the makers' average of 1 kW per cubic meter (Harvia, HUUM). Low end: the 1 kW per 50 cu ft shortcut, which is where Harvia's own worked example lands. Do not go below the low end.
Shop for a heater whose rated room range includes the calculated volume, ideally near the middle.
Room volume: 210 cu ft (5.95 m³). Glass and stone: 12 sq ft, counted as 47 cu ft of extra room (Harvia: 1.2 m³ per m²).
The same room by other published rules: HUUM 249 cu ft (7.1 m³); HUUM North American sheet 269 cu ft (7.6 m³).

Always confirm with the heater manufacturer and a licensed electrician. This is a rule-of-thumb estimate. It cannot see your insulation, vents, climate or wiring. The manual of the heater you buy has the final say on room volume, clearances, breaker and wire size.

The structured result is what calc() of the calculator returns, plus confirm and sources (links to the Harvia and HUUM pages each rule comes from). Rooms outside what the calculator covers (a side under 3 ft or over 19 ft 8 in, a ceiling under 5 ft 7 in or over 9 ft 10 in, more glass than the room has surface) come back as a tool error that says which number to check. Every rule and its source is in the calculator's README. Infrared cabins are sized differently and are not covered.

lookup_tradition

Argument

Values

Default

query

a name, local name or country: banya, savusauna, Japan, hamam. Case, accents and small typos are forgiven

none: lists all 28

list

true to list the names instead of searching

false

limit

most full records to return, 1 to 10

3

Each match is the full dataset row (heat source, typical air or water temperature in °F and °C, humidity, typical length of a stay, water on the stones, rituals, cooling, headwear, UNESCO status, notes) with the matching rows of the headwear table attached. The numbers are compiled reference ranges from general published descriptions, not measurements and not recommendations; an empty field means "not stated".

lookup_term

Argument

Values

Default

query

the word in any script (löyly, веник, 찜질방), its transliteration (venik), or an English meaning (whisk)

none

language

a language name or ISO code: Finnish, fi, ru

all 25

category

place, room, heat source, steam, tool, practice, person, headwear, cooling, expression, food and drink

all

limit

most terms to return, 1 to 25

8

With no query it lists the terms of the language or category given, or, with nothing at all, the languages and categories. The glossary is a reference word list, not a complete dictionary.

How results are returned

Each successful call returns structuredContent (matching the tool's outputSchema) and two text blocks: a plain-English answer, then the same structured data as JSON for clients that only read text. Arguments that do not fit the schema, and rooms the calculator refuses, come back as a tool result with isError: true and a message the model can act on. An unknown tool name is a protocol error.

Protocol

The server is built on the official TypeScript SDK, @modelcontextprotocol/server 2.x, and speaks both generations of MCP over stdio:

  • the current revision, 2026-07-28: stateless requests that carry their protocol version in _meta, and server/discover;

  • the earlier revisions that open with an initialize handshake (2025-11-25, 2025-06-18, 2025-03-26, 2024-11-05).

Which one is used is decided by how the client opens the connection. Standard output carries protocol messages only.

Tests

npm install
npm test
  • test/logic.test.mjs calls the tool logic directly: the vendored rules and data are the unchanged originals (SHA-256), the planner still gives the golden hash of sauna-planner-core over 42,534 plans, plan_session and size_heater return exactly what the original functions return, the makers' worked examples come out right, and the searches find what they should.

  • test/stdio-raw.test.mjs starts the server as a child process and talks to it in raw JSON-RPC with no SDK on the client side, once in the 2026-07-28 style and once per older protocol version: discovery or handshake, tools/list, 22 valid and failing tool calls compared with direct calls of the logic, 8 calls the input schema must refuse, an unknown tool, a missing _meta, an unsupported version, a clean exit when stdin closes, and nothing but protocol messages on stdout.

  • test/sdk-clients.test.mjs runs the same calls through the official clients: @modelcontextprotocol/client 2.x (automatic negotiation, pinned to 2026-07-28, and legacy mode) and @modelcontextprotocol/sdk 1.x. They are dev dependencies; without them these tests are skipped.

Files

File

What it is

server.mjs

The MCP server: tool names, descriptions, schemas, stdio

lib/tools.mjs

The tool logic as plain functions, and the search

vendor/

Unchanged copies of the planner and heater rules, with checksums

data/

Copies of the two datasets (CC BY 4.0)

test/

Tests (not in the npm package)

Credit and license

Made by DIVELUX, handmade wool sauna hats. More free sauna tools: https://diveluxsaunahat.com/pages/sauna-tools

Code: MIT License, © 2026 DIVELUX Inc. See LICENSE. The data in data/ is CC BY 4.0, © 2026 DIVELUX; see data/README.md.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      A
      quality
      C
      maintenance
      Enables AI agents to perform exact, reproducible mechanical computation — text diffing, hashing, JSON querying, calendar arithmetic, deadline-bounded regex extraction, CSV parsing, similarity scoring, unit and base conversion — instead of relying on probabilistic guessing. Runs dependency-free, read-only and offline over stdio, with schema validation and hard caps on every operation so no call can hang the client.
      15
      MIT
    • A
      license
      A
      quality
      B
      maintenance
      Gives AI assistants a read-only, offline interface to the FinOps Foundation's published guidance — its principles, phases, domains, 22 capabilities with Crawl/Walk/Run maturity assessments, personas, and an 88-entry KPI library — so FinOps questions are answered with the Foundation's own text and formulas rather than improvised ones. It exposes 11 tools, document resources, and four guided prompts over stdio, with every response carrying its source URL and license.
      11
      46 npm
      MIT
    • A
      license
      Not graded
      quality
      B
      maintenance
      Exposes four offline stdio tools that let local AI agents convert single values or in-process batches while rejecting ambiguous units and dimensional mismatches, and inspect or list canonical units with citations. All computation runs locally with zero network requests, structured machine-readable errors, and enforced item and payload limits.
      2
      MIT
    • A
      license
      Not graded
      quality
      C
      maintenance
      Exposes eighteen garden-planning tools to AI agents over stdio, enabling crop calendar and planting-window lookups, companion planting graphs, crop rotation and succession planning, pest and disease intelligence with beneficial-insect cross-references, and USDA zone plus climate classification. All data is curated and served offline, so agents can answer climate-aware scheduling questions without live API calls.
      1
      MIT