negativemeta/weeb-alexandria

Weeb Alexandria

About the project

Weeb Alexandria is a project that unifies anime, otaku, weeb, NSFW, and SFW knowledge into a single, simple, locally hosted knowledge base.

It gives local AI agents access to reliable information about the topics you ask them about, helping them retrieve grounded data and reduce hallucinations in this area of knowledge.

The project combines tag definitions, aliases, implications, characters, franchises, artists, traits, sources, and other useful information for anime-related research and image generation.

Current snapshot

The current tag_library.db snapshot contains approximately:

• 1.42 million unique tags
• 1.51 million tag records across sources
• 673,000 wiki entries
• 592,000 wiki entries with definitions
• 38,000 tag aliases
• 25,000 active aliases
• 24,000 tag implications
• 21,000 active implications

Database size: approximately 866 MB.

Snapshot date: 2026-08-25.

These counts refer to the published database snapshot and may change between releases.

Quick Start

1. Get the project

Clone the repository or download its source files into a local directory:

git clone https://github.com/NegativeMeta/weeb-alexandria.git "<WEEB_ALEXANDRIA_DIR>"

Alternative: without the Git CLI

Open the GitHub repository in a browser:

https://github.com/NegativeMeta/weeb-alexandria

Select Code → Download ZIP, extract the archive, and use the extracted folder as <WEEB_ALEXANDRIA_DIR>.

2. Download the database

Download the public database snapshot from Hugging Face:

hf download negativemeta/weeb-alexandria tag_library.db \
  --repo-type dataset \
  --local-dir "<WEEB_ALEXANDRIA_DIR>"

Alternative: without the Hugging Face CLI

Open the dataset in a browser:

https://huggingface.co/datasets/negativemeta/weeb-alexandria

Open Files and versions, download tag_library.db, and place it in <WEEB_ALEXANDRIA_DIR>.

The database must be located at:

<WEEB_ALEXANDRIA_DIR>\tag_library.db

2.1 Build the character-context index (recommended)

The derived index helps resolve queries such as Rika Higurashi by separating the character name from the franchise context. It also records canonical work relations such as:

furude_rika → higurashi_no_naku_koro_ni

Build or rebuild it with:

.venv/Scripts/python.exe scripts/build_context_index.py

It creates data/character_context.sqlite with the character_context, character_work_context, and context_index_metadata tables. The index is local/generated, intentionally excluded from Git and this dataset upload, and should be rebuilt after replacing tag_library.db. Metadata records the source database size and index schema version.

3. Connect it to Hermes

Register the local stdio MCP server:

hermes mcp add weeb-alexandria \
  --command "C:\\Windows\\System32\\cmd.exe" \
  --args /d /c "<WEEB_ALEXANDRIA_DIR>\\run.bat"

4. Connect it to LM Studio

Open Program → Install → Edit mcp.json and add:

{
  "mcpServers": {
    "weeb-alexandria": {
      "command": "C:\\Windows\\System32\\cmd.exe",
      "args": [
        "/d",
        "/c",
        "<WEEB_ALEXANDRIA_DIR>\\run.bat"
      ]
    }
  }
}

Start a new chat and confirm that the five Weeb Alexandria tools are available.

5. Try it in a conversation

You do not need to call the MCP tools manually. Ask the model naturally, for example:

Tell me what Inugami Korone looks like. You can use Weeb Alexandria to check the character information and relevant tags.

Credits, provenance, and licensing

See CREDITS.md for source attribution and the redistribution notice. The repository MIT license applies to project software and documentation only to the extent that the project can license those materials; it is not a blanket license for all records in this dataset. Upstream terms and source-specific attribution requirements remain applicable. Synthetic definitions marked with lang='llm' are project-generated additions.

The derived character-context index is a local build artifact and is not included in this dataset upload. Rebuild it from tag_library.db with scripts/build_context_index.py when needed.

Structure

WeebAlexandria/
├── weeb_alexandria_mcp/   Active MCP server
├── .venv/                  Python runtime for the MCP
├── tag_library.db          Main unified knowledge base
├── raw/                    Downloaded source data
│   ├── animadex/           Original AnimaDex database
│   ├── danbooru/
│   ├── e621/               Processed wiki and tag data
│   ├── gelbooru/
│   └── danbooru_wiki_extra/
├── reports/                Audits and review lists
├── scripts/                Data maintenance and fusion scripts
├── data/character_context.sqlite
│                           Derived character-context index (local)
├── data/backups/           Database backups
├── CREDITS.md              Credits and source information
├── run.bat                 MCP launcher
└── README.md

The original AnimaDex web application, legacy MCP, and visual assets are not part of the active runtime. They are preserved reversibly at:

C:\Users\johin\Code_Library\AI\WeebAlexandria_legacy_archive

Databases

tag_library.db contains:

• Tags and categories.
• Danbooru, e621, and Gelbooru definitions.
• Aliases and implications.
• Synthetic definitions marked with lang='llm'.
• Owned structured character profiles and trait mappings (character_profiles, trait_definitions, and character_traits).
• Artist and copyright discovery from the global tags table.

raw/animadex/animadex.db is preserved under its original name for audit and seed recovery only. The active MCP does not open it and does not require any legacy structured tables.

The historical structured seed and migration record are documented in the repository migration record. search_knowledge returns structured results under the entities namespace.

MCP

Launcher:

run.bat

Available tools:

• search_knowledge
• get_tag_knowledge
• search_characters
• get_character
• get_sources_status

Weeb Alexandria performs its queries locally and does not require the original AnimaDex Flask server.

Glossary

See GLOSSARY.md for a simple explanation of the database contents and MCP tools. Spanish, Simplified Chinese, and Japanese versions are available in GLOSSARY.es.md, GLOSSARY.zh-CN.md, and GLOSSARY.ja.md.

Credits and sources

• AnimaDex — historical source of the small structured seed; retained for attribution and migration provenance, not as an active runtime dependency.
• Danbooru — wiki, tag metadata, aliases, implications, and popularity data.
• e621 — wiki and tag metadata.
• Gelbooru — wiki and tag metadata.
• CREDITS.md — full acknowledgements, attribution, and source details.