kip-mcp-server

An MCP server that lets an AI assistant (such as Claude, Codex, or Gemini) look at your boat's Signal K data and help you design and install KIP dashboards — the gauges and panels you see on your chartplotter or tablet.

You stay in control: the assistant shows you a picture of each dashboard first, and it only saves anything to your boat after you say yes.

What it does

• Looks at your boat's data. It finds the values your boat reports through Signal K — speed, wind, depth, batteries, engine, and so on.

• Suggests dashboards. It proposes a general dashboard plus ones for specific jobs: sailing, motoring, power, anchoring, navigation, and weather.

• Shows you a preview. You see a simple picture of each dashboard before anything is saved.

• Installs them — with your OK. It writes the dashboards to your KIP setup, or hands you a file you can import yourself.

Related MCP server: KiCad MCP Server

What it looks like

A short session with an AI assistant, after pointing it at your boat:

You:  Look at my boat's data and suggest some KIP dashboards.

  → analyze_signalk_data            142 paths · speed, wind, depth, battery, engine
  → recommend_dashboard_set         6 dashboards the data supports

Assistant: Your boat reports navigation, wind, depth, electrical and engine data,
so I can build these. Here's the "Sailing" one — a preview, nothing saved yet:

  ┌─────────────┬─────────────┬───────────────────────────┐
  │ SOG  6.2 kn │ AWS 14 kn   │                           │
  ├─────────────┼─────────────┤        wind steering       │
  │ DPT  8.4 m  │ AWA  42°    │                           │
  └─────────────┴─────────────┴───────────────────────────┘

You:  Looks good — install the sailing and anchor ones.

  → apply_kip_config (dry run)      2 dashboards → KipConfig (global)
Assistant: This will add "Sailing" and "Anchor" to KIP. Go ahead?
You:  Yes.
  → apply_kip_config                written ✓  — open KIP to see them.

The assistant always shows a preview first and asks before writing anything to your boat.

Quick start (for boat owners)

You need a Signal K server running on your boat (or on your network), with KIP installed.

1. Find your Signal K address. It usually looks like http://your-boat:3000. Note the host name (or IP) and port.

2. Get this server. You don't need to install anything by hand — your AI assistant can run it on demand with npx (which comes with Node.js 24 or newer):

   npx -y kip-mcp-server

   The next step wires this command into your assistant. (Prefer to build from source? See Develop below.)

3. Connect it to your AI assistant. Pick your assistant in docs/clients and follow the short setup there. You tell the assistant your boat's host and port.

4. Ask it to help. Say something like "Look at my boat's data and suggest some KIP dashboards." Review the previews it shows you.

5. Say yes. When you're happy, tell it to go ahead. It asks before writing anything. If your Signal K server is older, it gives you a KipConfig.json file to import from KIP's Settings instead.

How to connect it

The server reads a few settings from its environment:

Reading your data needs no login. Writing dashboards to the server needs either a token or a username and password; you can always use the file-export option instead, which needs nothing extra. For how to get a token (or why a username/password is simpler), see Signal K authentication.

Remote access over HTTP (optional, advanced)

By default the server talks over stdio — the assistant runs it as a local subprocess. There is also an optional HTTP mode (kip-mcp-http) for hosting the server so a remote assistant (such as Claude.ai) can reach it over the network. It is opt-in and meant for a single operator behind a reverse proxy that adds TLS.

# Behind a TLS-terminating reverse proxy that forwards to 127.0.0.1:3017
MCP_BEARER_TOKEN=a-long-random-secret \
MCP_PUBLIC_URL=https://boat.example.com/mcp \
  npx kip-mcp-http

Every request must present a valid Authorization: Bearer <token> and pass a Host/Origin allowlist before it reaches the MCP layer. The server refuses to start if it would bind to a non-loopback address or run without a bearer token, unless you set MCP_ALLOW_INSECURE=true. It uses one Signal K login for all sessions, so treat it as a single-tenant deployment; the same SIGNALK_* settings above apply.

What the assistant can do

The server gives the assistant a set of tools, grouped by job:

• Look at the boat — list the data paths, their units, and which plugins are installed.

• Know KIP's parts — list every KIP widget and how it's configured, plus the colours, icons and units KIP understands.

• Design dashboards — suggest a set of dashboards, build one for a chosen job, and draw a preview.

• Review the UX — critique a dashboard for marine usability (hierarchy, grouping, consistency, labels) with severity-tagged findings; the review_dashboard prompt drives it, backed by a deterministic check_dashboard_ux lint.

• Check and save — check a dashboard is well-formed, export it to a file, or write it to the boat (asking first).

More help

• Recipes — worked examples of asking the assistant to design dashboards.

• Troubleshooting — fixes for the common problems, starting with the built-in --doctor check.

• Tool reference — every tool and prompt the assistant can use.

Glossary

A few terms, in plain words:

• Signal K — the open system many boats use to share data (speed, wind, depth, …) over the network.

• path — the name of one piece of data, like navigation.speedOverGround (speed over ground).

• KIP — the app that shows your boat's data as dashboards of gauges and panels.

• widget — one gauge or panel on a dashboard (a number, a dial, a wind display, …).

• dashboard — a screen full of widgets, laid out on a grid.

• token — like a password for software: it lets the server save changes to your boat.

• dry run — the assistant tells you what it would do, without actually doing it. Saving is a dry run by default.

Develop

This project uses Node.js 24 (LTS). Common commands:

npm install        # install dependencies
npm run typecheck  # check types
npm run lint       # check code style
npm test           # run the tests
npm run build      # compile to dist/
npm run smoke      # start the built server and check it answers
npm run ci         # run the full set of checks

For a full guide to running and testing the server locally during development — pointing an AI client or the MCP Inspector at a local build, running the HTTP transport, and testing against Signal K (or offline) — see docs/development.md.

Commits follow Conventional Commits; releases and version numbers are produced automatically from those commit messages.

License

MIT — see LICENSE.