kaedemeta-mcp-server
# KaedeMeta MCP Server
**High-concurrency lossless media & structured metadata extraction MCP Server for AI Agents.**
Give Claude Desktop, Cursor, Windsurf, or any MCP-compatible agent the ability to resolve original uncompressed image URLs, highest-bitrate MP4 streams, multi-page manga sets, and Ugoira archives from **X (Twitter)**, **Pixiv**, and **Danbooru** — with zero local cookie maintenance.
- **Web Workbench & Instant API Keys:** [https://kaedemeta.com](https://kaedemeta.com)
- **Free Trial:** Claim **10 free credits** instantly on the workbench (no credit card required).
---
## Key Capabilities
- **Lossless Original Resolution:** Automatically upgrades media URLs to `:orig` / uncompressed master files and selects maximum-bitrate video variants.
- **Batch Creator & Tag Feeds:** Extract up to 24 posts per call from `@username` timelines, Pixiv creator portfolios, or Danbooru tag queries.
- **Zero-State Execution:** No media binaries or user session credentials are ever stored on our servers.
- **Success-Only Metering:** 0 credits deducted on upstream 4xx/5xx errors or empty feeds.
---
## Quick Setup
### Option A: Native Streamable HTTP / JSON-RPC (Recommended)
For clients supporting remote HTTP MCP endpoints (Cursor, Claude Desktop, OpenClaw, custom agents):
```json
{
"mcpServers": {
"kaedemeta": {
"url": "https://api.kaedemeta.com/mcp?api_key=YOUR_KAEDEMETA_API_KEY"
}
}
}
```
### Option B: Local `uvx` Stdio Bridge
If your MCP host requires a local `stdio` command, run the zero-config Python bridge:
```json
"mcpServers": {
"kaedemeta": {
"command": "uvx",
"args": ["kaedemeta-mcp"],
"env": {
"KAEDEMETA_API_KEY": "YOUR_KAEDEMETA_API_KEY"
}
}
}
```
---
## Exposed MCP Tools
| Tool Name | Description |
| :--- | :--- |
| `extract_post_metadata` | Extract structured metadata and original high-resolution media URLs from a single X, Pixiv, or Danbooru post URL/ID. |
| `extract_creator_feed` | Batch extract original media URLs and metadata from a creator profile (`@username`, `users/ID`) or Danbooru tag query. |
---
## Example Prompts for Your AI Agent
- *"Fetch the latest 24 original artworks from `[https://www.pixiv.net/users/2188539](https://www.pixiv.net/users/2188539)` and summarize the tags and publishing dates."*
- *"Extract the uncompressed original images from `[https://x.com/gM4zFiJCWbz3G4y](https://x.com/gM4zFiJCWbz3G4y)`."*
- *"Query Danbooru for `original 1girl solo` and give me the direct high-res file URLs."*
---
## License
MIT License. Official MCP Server implementation for the KaedeMeta Core API ([kaedemeta.com](https://kaedemeta.com)).
TDQS
Scored across 2 tools
The two tools have clearly distinct scopes: one handles a single post, while the other batch-extracts from a creator feed or tag listing. An agent can easily choose between single-item and batch/feed operations.
Both tool names follow the same extract_* snake_case convention with clear noun phrases. There is no mixing of styles or vague verbs.
Two tools is on the thin side for a metadata extraction server, though the pair covers single-post and batch-feed workflows. It is borderline rather than clearly well-scoped because additional query or lookup operations might reasonably belong here.
The surface covers the core extraction lifecycle: one post and creator/tag feeds across the named platforms. Minor gaps exist, such as keyword search or multi-post URL handling, but agents can work around them for the stated purpose.