Skip to main content
Glama
README.md
# zerochan-mcp

[![Python 3.10+](https://img.shields.io/badge/python-3.10%2B-blue.svg)](https://www.python.org/)
[![License: MIT](https://img.shields.io/badge/License-MIT-green.svg)](LICENSE)
[![MCP](https://img.shields.io/badge/MCP-compatible-blueviolet)](https://modelcontextprotocol.io/)

<a href="https://glama.ai/mcp/servers/citronlegacy/zero-chan-mcp">
  <img width="380" height="200" src="https://glama.ai/mcp/servers/citronlegacy/zero-chan-mcp/badge" alt="DeepGHS MCP server rating" />
</a>

A Python [MCP](https://modelcontextprotocol.io/) server wrapping the read-only [Zerochan](https://www.zerochan.net) anime image board API. Connect it to Claude Desktop, Cursor, or any MCP-compatible AI assistant to browse, search, and inspect one of the web's most comprehensively tagged anime image collections — directly from a conversation.

Built as a sister server to [gelbooru-mcp](https://github.com/citronlegacy/gelbooru-mcp).

---

## ✨ Features

- **Browse** the global Zerochan feed — sort by newest or most-favorited, filter by dimensions or dominant color
- **Search** by one or more tags simultaneously using Zerochan's natural Title Case format (`Hatsune Miku`, not `hatsune_miku`)
- **Strict mode** — narrow results to entries where a specific tag is the *primary* subject, cutting group shots and incidental appearances
- **Full entry details** — all tags, full/medium/small image URLs, source, dimensions, favorites, and anime/manga/game associations
- Returns results as a clean **Markdown table** or raw **JSON** — your choice
- Fully **Pydantic-validated** inputs with enum constraints and custom tag validators
- Clear, actionable **error messages** for rate limits, auth failures, and timeouts

---

## 📦 Installation

### Prerequisites

- Python 3.10+
- `git`

### Quick Start

```bash
git clone https://github.com/citronlegacy/zerochan-mcp.git
cd zerochan-mcp
chmod +x setup.sh && ./setup.sh
```

Or without chmod:

```bash
bash setup.sh
```

### Manual Install

```bash
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
```

---

## 🔑 Authentication

Zerochan requires a valid **username** in every request's User-Agent header. Unauthenticated requests are blocked.

```bash
export ZEROCHAN_USERNAME="YourZerochanUsername"
```

> Your username is included in the User-Agent string as `zerochan-mcp - YourUsername`, exactly as Zerochan's API requires. The server validates this on every call and returns a clear error message if it is missing — no silent failures.

---

## ▶️ Running the Server

```bash
# via the venv created by setup.sh:
.venv/bin/python server.py

# or with the venv activated:
source .venv/bin/activate
python server.py
```

---

## ⚙️ Configuration

### Claude Desktop

Add the following to your `claude_desktop_config.json`:

```json
{
  "mcpServers": {
    "zerochan-mcp": {
      "command": "/absolute/path/to/zerochan-mcp/.venv/bin/python",
      "args": ["/absolute/path/to/zerochan-mcp/server.py"],
      "env": {
        "ZEROCHAN_USERNAME": "YourZerochanUsername"
      }
    }
  }
}
```

### VS Code / Cursor / Other MCP Clients

Configure according to your client's documentation:

- **Command:** `/absolute/path/to/zerochan-mcp/.venv/bin/python`
- **Args:** `/absolute/path/to/zerochan-mcp/server.py`
- **Transport:** stdio
- **Env:** `ZEROCHAN_USERNAME=YourZerochanUsername`

### Test with MCP Inspector

```bash
npx @modelcontextprotocol/inspector .venv/bin/python server.py
```

---

## 💡 Usage Examples

### Browse the latest anime art

> "Show me the 10 most recent images on Zerochan."

The LLM calls `zerochan_browse` with `limit: 10, sort: "id"` and returns a paginated Markdown table of the newest uploads — IDs, tags, dimensions, and direct links.

---

### Find the all-time most favorited images

> "What are the most popular images on Zerochan of all time?"

The LLM calls `zerochan_browse` with `sort: "fav", time_range: "0"` and returns the highest-favorited entries globally.

---

### Search for a character

> "Find me portrait-mode images of Hatsune Miku."

The LLM calls `zerochan_search` with `tags: ["Hatsune Miku"], dimensions: "portrait"`. No tag normalization required — Zerochan's Title Case maps almost one-to-one with how people naturally write character names.

---

### Multi-tag search — character + theme

> "Show me Rem from Re:Zero with an umbrella."

The LLM calls `zerochan_search` with `tags: ["Rem", "Umbrella"]`. Zerochan joins them as `/Rem,Umbrella?json` internally, returning only images tagged with **both** simultaneously.

---

### Strict mode — images where a character is the main subject

> "I only want images where Hatsune Miku is the actual primary subject, not just tagged."

The LLM calls `zerochan_search` with `tags: ["Hatsune Miku"], strict: true`. This filters to entries where Hatsune Miku is the **primary tag**, removing group shots and background appearances.

---

### Get full metadata for a specific post

> "Give me all the details on Zerochan entry 3793685 — tags, source, full image URL."

The LLM calls `zerochan_get_entry` with `entry_id: 3793685` and returns a complete metadata card: every tag, full/medium/small image URLs, source, dimensions, favorites, and associated anime/game/manga categories.

---

### Raw JSON for downstream processing

> "Search for Yotsuba images and give me the raw JSON."

The LLM calls `zerochan_search` with `tags: ["Yotsuba"], response_format: "json"` and returns the unprocessed Zerochan API payload ready for further processing.

---

## 🛠️ Available Tools

| Tool | Description | Key Parameters |
|---|---|---|
| `zerochan_browse` | Browse the global Zerochan feed | `page`, `limit`, `sort`, `time_range`, `dimensions`, `color` |
| `zerochan_search` | Search entries by one or more tags | `tags`, `strict`, `page`, `limit`, `sort`, `dimensions`, `color` |
| `zerochan_get_entry` | Get full metadata for a single entry by ID | `entry_id` |

All tools accept `response_format: "markdown"` (default) or `"json"`.

---

## 📖 Tools Reference

### `zerochan_browse`

Browse all Zerochan entries with optional filtering and pagination. No tag required — queries the full global feed.

**Parameters**

| Parameter | Type | Required | Default | Description |
|---|---|---|---|---|
| `page` | int | ❌ | 1 | Page number (starts at 1) |
| `limit` | int | ❌ | 20 | Results per page (1–250) |
| `sort` | enum | ❌ | `id` | `id` = newest first, `fav` = most favorited |
| `time_range` | enum | ❌ | — | `0` = all time, `1` = last 7 000 entries, `2` = last 15 000 entries |
| `dimensions` | enum | ❌ | — | `large`, `huge`, `landscape`, `portrait`, `square` |
| `color` | str | ❌ | — | Dominant color name, e.g. `blue`, `red`, `pink` |
| `response_format` | enum | ❌ | `markdown` | `markdown` or `json` |

**Example response**

```
### Zerochan Results from `Global Feed`

| ID | Tags | Dimensions | Favorites | Full Image |
|---|---|---|---|---|
| 4666171 | Female, Twin Tails, Flower, Hatsune Miku, Music (+33 more) | 1000×1500 | ? | View |
```

---

### `zerochan_search`

Search Zerochan entries by one or more tags. Tags use Zerochan's Title Case with spaces format.

**Parameters**

| Parameter | Type | Required | Default | Description |
|---|---|---|---|---|
| `tags` | list[str] | ✅ | — | One or more tags, e.g. `["Hatsune Miku"]` or `["Rem", "Umbrella"]` |
| `strict` | bool | ❌ | false | Restrict to entries where the **first** tag is the primary tag (single-tag only) |
| `page` | int | ❌ | 1 | Page number |
| `limit` | int | ❌ | 20 | Results per page (1–250) |
| `sort` | enum | ❌ | `id` | `id` = newest, `fav` = most favorited |
| `dimensions` | enum | ❌ | — | `large`, `huge`, `landscape`, `portrait`, `square` |
| `color` | str | ❌ | — | Dominant color filter |
| `response_format` | enum | ❌ | `markdown` | `markdown` or `json` |

> **Tag format:** Zerochan uses Title Case with spaces — `Hatsune Miku` not `hatsune_miku`. Natural-language character names work verbatim in most cases.

> **Multi-tag:** `tags: ["Hatsune Miku", "Flower"]` → `/Hatsune+Miku,Flower?json`

> **Strict mode:** Only works with a single tag. Silently ignored when multiple tags are provided.

---

### `zerochan_get_entry`

Retrieve complete metadata for a single Zerochan post by its numeric ID.

**Parameters**

| Parameter | Type | Required | Default | Description |
|---|---|---|---|---|
| `entry_id` | int | ✅ | — | Numeric post ID, e.g. `3793685` (visible in the post URL) |
| `response_format` | enum | ❌ | `markdown` | `markdown` or `json` |

**Example response (markdown)**

```
## Zerochan Entry #3793685

**Primary Tag:** Hatsune Miku
**Dimensions:** 1748 × 2480
**Favorites:** 42
**Source:** https://...
**Tags (28):** Hatsune Miku, VOCALOID, Female, Twin Tails, ...

**Full Image:** https://static.zerochan.net/...
**Medium Preview:** https://static.zerochan.net/...
```

**Response schema (JSON)**

```json
{
  "id": 3793685,
  "primary": "Hatsune Miku",
  "tags": ["Hatsune Miku", "VOCALOID", "Female", "..."],
  "width": 1748,
  "height": 2480,
  "fav": 42,
  "source": "https://...",
  "full": "https://static.zerochan.net/...",
  "medium": "https://static.zerochan.net/...",
  "small": "https://static.zerochan.net/...",
  "anime": "VOCALOID",
  "manga": null,
  "game": null
}
```

---

## 🤖 Notes for LLMs

- **Tag format:** Zerochan uses Title Case with spaces — `Hatsune Miku`, `Rem`, `Attack On Titan`. Unlike Gelbooru/Danbooru, natural-language names map directly to Zerochan tags in most cases.
- **Strict mode:** Use it when the user asks specifically for images *of* a character, not just images *featuring* them. Only valid with a single tag.
- **Multi-tag search:** Each entry must match **all** tags. Use for character + theme combos (`["Rem", "Rain"]`, `["Naruto", "Ramen"]`).
- **Pagination:** `zerochan_browse` and `zerochan_search` return up to 250 results per call. Use `page` to paginate through up to 100 pages (20 000 entries) per tag.
- **Rate limit:** 60 requests/minute. Avoid tight loops; space calls when paginating deeply.
- **Entry IDs:** Visible in Zerochan URLs (`zerochan.net/3793685`). Pass them to `zerochan_get_entry` to resolve full metadata.

---

## ⚠️ Known Limitations

- **Rate limit:** 60 requests/minute enforced server-side. Exceeding this may trigger a temporary ban.
- **Username required:** Every request must carry a valid Zerochan username in the User-Agent. The server validates this and returns a clear error if missing.
- **Strict mode + multi-tag:** Strict mode is silently ignored when more than one tag is provided — Zerochan API limitation.
- **No write access:** Zerochan's public API is fully read-only. Uploading, favoriting, and commenting are not supported.
- **No tag autocomplete:** Zerochan does not expose a tag search/autocomplete endpoint in the JSON API. Use the website for tag discovery when needed.

---

## 🐛 Troubleshooting

**`ZEROCHAN_USERNAME is not set`**
- Add it to your client's `"env"` block, or: `export ZEROCHAN_USERNAME="YourUsername"`

**Error 403 — Access denied**
- Ensure your Zerochan username is correct and your account is in good standing.
- Blank or missing User-Agent strings are rejected by Zerochan.

**Error 429 — Rate limit**
- Wait 60 seconds and retry. Avoid paginating in rapid loops.

**Error 404 — Not found**
- The tag or entry ID does not exist on Zerochan. Check spelling; entry IDs come from post URLs.

**Strict mode returns fewer results than expected**
- By design — strict mode returns only entries where the tag is the **primary** classification. Use non-strict for broader results.

---

## 🤝 Contributing

Pull requests are welcome! If you find an API edge case not handled, a tag format inconsistency, or want to add a new filter, open an issue or PR.

### Development Setup

```bash
git clone https://github.com/citronlegacy/zerochan-mcp.git
cd zerochan-mcp
bash setup.sh
source .venv/bin/activate
```

---

## 📄 License

MIT License — see [LICENSE](LICENSE) for details.

---

## 🔗 Links

- 🌐 [Zerochan](https://www.zerochan.net)
- 🔧 [MCP documentation](https://modelcontextprotocol.io/)
- 🗂️ [MCP Registry](https://registry.modelcontextprotocol.io)
- 🔍 [Glama MCP Directory](https://glama.ai/mcp/servers)
- 🎨 [gelbooru-mcp](https://github.com/citronlegacy/gelbooru-mcp) — sister MCP server
- 🐛 [Bug Reports](https://github.com/citronlegacy/zerochan-mcp/issues)
- 💡 [Discussions](https://github.com/citronlegacy/zerochan-mcp/discussions)

---

## Relation to MultiBoru MCP

This server is one piece of a planned **MultiBoru** federated image board MCP that will aggregate Gelbooru, Danbooru, and Zerochan queries in parallel with cross-site tag normalization.

| Site | Max Tags | Tag Format | Auth |
|---|---|---|---|
| Gelbooru | Many | `snake_case` | Optional API key |
| Danbooru | 2 (free) / 6 (Gold) | `snake_case` | Optional |
| Zerochan | Many | `Title Case` | Username in User-Agent |

TDQS

A4.5/5.0

Scored across 3 tools

Disambiguation5/5

Each tool has a clearly distinct purpose with no overlap: zerochan_browse retrieves the global feed, zerochan_get_entry fetches a single entry by ID, and zerochan_search finds entries by tags. The descriptions explicitly differentiate their use cases, making it impossible for an agent to confuse them.

Naming Consistency5/5

All three tools follow a perfect verb_noun pattern with the prefix 'zerochan_' and consistent snake_case: zerochan_browse, zerochan_get_entry, and zerochan_search. The naming is predictable and uniform throughout the set.

Tool Count4/5

Three tools is reasonable for a ZeroChan image browsing server, covering core operations (browse, get, search). However, it feels slightly thin as it lacks tools for user-specific actions like favorites or uploads, which might be expected in a full-featured image platform integration.

Completeness4/5

The toolset covers essential read operations for the ZeroChan domain: browsing, searching, and retrieving entries. Minor gaps exist, such as no tools for user authentication, managing favorites, or uploading content, but agents can still perform basic image discovery and viewing tasks effectively.

Maintenance

ActivityInactive
ResponsivenessNo issues