⚓ imo-rules-mcp

Bring your corpus. Keep the rules local.

A read-only MCP server for offline semantic search, keyword search, document retrieval, citation traversal, and structured browsing across a locally supplied IMO regulatory corpus.

Tools · Set up a corpus · Rights boundary · Full technical reference

The server is open. The corpus is yours to license, store, and protect.

imo-rules-mcp exposes a local corpus of maritime instruments to MCP-compatible clients without crawling the web or redistributing regulation text.

MCP tools

Tool	Purpose
`imo_semantic_search`	Meaning-based retrieval with ranked regulations, breadcrumbs, snippets, and document keys.
`imo_keyword_search`	Literal title or body search.
`imo_get_document`	Retrieve one normalized regulation with navigation context and citations.
`imo_list_instruments` / `imo_browse`	Navigate instruments, chapters, and regulations.
`imo_get_citations`	Follow outbound cross-references.
`imo_stats`	Inspect corpus coverage and index state.

Architecture

flowchart LR
    A["Licensed local corpus"] --> B["SQLite documents"]
    A --> C["TOC tree"]
    A --> D["Local embedding index"]
    B --> E["MCP server"]
    C --> E
    D --> E
    E --> F["Offline MCP client"]

Quickstart

Install:

npm install

Import a compatible exported corpus:

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

Start the stdio server:

npm start

A repository-scoped .mcp.json is included for hosts that support project-local MCP configuration.

Corpus and rights boundary

This repository contains server code only. It does not ship SOLAS, MARPOL, STCW, IMO Codes, or any other regulation corpus.

Supply only material you are licensed to use.
Keep corpus/ outside version control.
Do not publish or redistribute proprietary regulation text.
If a convenience bundle includes corpus data, keep that repository private and apply the applicable rights controls.
Server-code licensing does not grant rights to the corpus.

Runtime contract

Read-only tools only.
No crawler and no required network path.
Semantic search loads its local model/index on demand.
Other tools operate directly on the local corpus structures.
Corpus completeness, currency, interpretation, and legal authority remain external responsibilities.

Full technical reference

The original detailed README — including the expected corpus layout, environment variables, client configuration, and tool-by-tool behavior — is preserved unchanged at:

README.technical.2026-08-25.md

License

Server code is MIT. Corpus data is excluded from that grant.