imo-rules-mcp

An MCP server that exposes a local corpus of IMO maritime rules and regulations (SOLAS, MARPOL, STCW, the codes, etc.) to MCP-compatible local clients for offline semantic search and reading. No crawler, no network; it serves a local corpus you provide.

Bring Your Own Corpus

This repository contains only the server code. It does not include any regulation text. The IMO rules and regulations corpus is copyrighted by its rights holders and is licensed to you separately. Do not commit it to a public repository or redistribute it. The corpus/ directory is gitignored for this reason. If you choose to bundle a corpus into a repo for convenience, keep that repository private.

Related MCP server: RAG In A Box MCP Server

What You Provide

A corpus directory (default ./corpus/) containing:

corpus/
  imo-corpus.sqlite     # SQLite corpus (documents, nodes, citations)
  toc.json              # navigation tree (instruments, chapters, regulations)
  rag/
    config.json
    meta.json
    vectors.f32         # one embedding per page
    models/             # local embedding model

This layout is produced by the companion KRcrawl exporter (export:full-help

• rag:index). Import it with:

npm install
node scripts/import-corpus.mjs /path/to/KRcrawl

Or skip the copy and point the server at an existing build with env vars.

Tools

imo_semantic_search lazily loads the bundled local embedding model on first call; other tools are instant. The server is read-only and offline.

Use It

A project-scoped .mcp.json is included for MCP hosts that support repository-local configuration. For hosts that use a global config file, add a server entry like this and point cwd at this repo so ./corpus resolves:

{
  "mcpServers": {
    "imo-rules": {
      "command": "npx",
      "args": ["tsx", "src/server.ts"],
      "cwd": "/path/to/imo-rules-mcp"
    }
  }
}

Smoke-test standalone:

npm start

The process speaks MCP over stdio; press Ctrl-C to stop.

Configuration

To run against an existing KRcrawl build in place, set:

IMO_DB=<KRcrawl>/data/krcon.sqlite
IMO_RAG=<KRcrawl>/dist/help/full-clone/assets/rag
IMO_TOC=<KRcrawl>/dist/help/full-clone/toc.json

License

Server code: MIT (see LICENSE). The license covers the code only, not any corpus data you supply.