glossarygen

glossarygen

A project accumulates names faster than it accumulates explanations for them: the same concept shows up as user_id in the database layer, UserID in an API type, and "user id" in the docs, and nobody can say whether those are one term or three. glossarygen settles that kind of question from facts you already have. You supply term occurrences — each name exactly as you found it, where you found it, and a definition when the docs carry one — and glossarygen hands back a single merged glossary: one entry per distinct term, with every source the term was seen in and the definition the docs gave it, if any. It writes no definitions of its own and fetches nothing; it only normalizes and merges what you supplied.

Normalization follows a fixed recipe, in order. First, every run of underscores or hyphens becomes a single space. Then camel humps are split: a space goes before an uppercase letter that follows a lowercase letter or a digit, and before an uppercase letter that ends an acronym run and starts a capitalized word, so HTTPServer becomes HTTP Server. The result is split on whitespace and every word is lowercased; the words joined with single spaces form the term's key. That is why parseHTTPResponse keys as parse http response, and why user_id and UserID land on the same key, user id, and merge into one entry.

Each entry shows its term in display form — the key with its first character uppercased, so parse http response reads as Parse http response — together with the distinct sources across all its occurrences, sorted ascending, and how many occurrences merged into it. The entry's definition is the first one supplied in your order among the merged occurrences; an empty-string definition still counts as supplied and is kept exactly as given, and when no occurrence supplies one the definition is null. Entries come back sorted by key, and the counts report how many occurrences you supplied, how many distinct terms they merged into, and how many of those terms carry a definition versus not.

A name made only of separators, like ___, normalizes to no words at all and is a validation error naming its path, as are a missing or empty term list, a non-object occurrence, a missing, empty, or non-string name or source, and a definition that is present but not a string. Invalid input produces no partial glossary. Unknown extra fields are ignored.

CLI

npx @reneturcios/glossarygen build --input args.json
npx @reneturcios/glossarygen --version

args.json contains the full arguments object:

{
  "terms": [
    {
      "name": "user_id",
      "source": "src/user.ts"
    },
    {
      "name": "UserID",
      "source": "docs/api.md",
      "definition": "Stable identifier for a user account."
    },
    {
      "name": "parseHTTPResponse",
      "source": "src/http.ts"
    }
  ]
}

The glossary is printed as JSON:

{
  "entries": [
    {
      "term": "Parse http response",
      "sources": ["src/http.ts"],
      "occurrences": 1,
      "definition": null
    },
    {
      "term": "User id",
      "sources": ["docs/api.md", "src/user.ts"],
      "occurrences": 2,
      "definition": "Stable identifier for a user account."
    }
  ],
  "counts": {
    "supplied": 3,
    "distinct_terms": 2,
    "with_definition": 1,
    "without_definition": 1
  }
}

Exit codes: 0 on success, 1 on a runtime error (including an unreadable or invalid input file, or invalid input), 2 on a usage error.

MCP server

npx -y @reneturcios/glossarygen mcp

MCP client config:

{
  "mcpServers": {
    "glossarygen": {
      "command": "npx",
      "args": ["-y", "@reneturcios/glossarygen", "mcp"]
    }
  }
}

Tools

• build_glossary — Build a project glossary from caller-supplied terms: identifier and doc names are normalized and merged, each entry carrying its sources and its supplied definition, if any.

  • terms (array, required, at least 1 item) — the term occurrences to merge, in the caller's order; each item:
    • name (string, required, non-empty) — term as found: an identifier or a doc phrase
    • source (string, required, non-empty) — where the term was found (a path or link), echoed as the entry's source
    • definition (string, optional) — a definition found in the docs

  Returns {entries, counts}, where entries holds one {term, sources, occurrences, definition} per distinct normalized term, sorted by term key, and counts is {supplied, distinct_terms, with_definition, without_definition}.

Docker

docker run -i ghcr.io/cloudcorprecords/glossarygen
docker run --rm -v "$PWD:/w" ghcr.io/cloudcorprecords/glossarygen build --input /w/args.json

Development

npm install
npm run build   # tsc -> dist/
npm test        # tsx --test test/*.test.ts (pure local, no network)
npm run dev     # run the CLI from source

License

MIT — Copyright (c) 2026 Rene Turcios