filewash
Privacy-first file tools for AI agents, enabling operations like PDF merge/split, image compression/convert, metadata stripping, and background removal without storing files.
README
filewash — privacy-first file tools for AI agents
The file-processing layer for agents and apps that can't send files to a third party. Compress, convert, resize, merge/split PDFs, strip metadata, and remove backgrounds — exposed as an MCP server, a metered REST API, and an npm SDK. Zero-retention by default: files are processed in memory and never stored or logged.
Built from filewash.app's client-side WASM tools. Same logic, now callable by your agents.
Why this exists
Every other file-tools API and MCP server is a wrapper around an upload-to-server service. That's a non-starter for legal, healthcare, finance, and any GDPR-bound workflow — and it's a liability when you're piping documents into an LLM. filewash is the one built around not keeping your files, with two tools nobody else ships as an MCP: background removal and metadata scrubbing.
MCP server (Claude Desktop, Cursor, Claude Code)
// claude_desktop_config.json → mcpServers
{
"filewash": { "command": "node", "args": ["/path/to/filewash-api/src/mcp/server.js"] }
}
Your agent now has 10 tools. Ask it: "strip the location metadata from these photos before I upload them" or "merge these contracts and rotate the scanned pages."
REST API
# discover
curl https://api.filewash.app/v1/tools
# strip EXIF/GPS from a photo (returns the cleaned image)
curl -X POST https://api.filewash.app/v1/tools/image_strip_metadata \
-H "Authorization: Bearer $FILEWASH_KEY" \
-F 'files=@photo.jpg' -o clean.jpg
# merge PDFs
curl -X POST https://api.filewash.app/v1/tools/pdf_merge \
-H "Authorization: Bearer $FILEWASH_KEY" \
-F 'files=@a.pdf' -F 'files=@b.pdf' -o merged.pdf
Every response carries X-Filewash-Meta (operation result) and X-Filewash-Quota-Remaining.
Tools
| Tool | Category | What it does |
|---|---|---|
pdf_merge |
Merge PDFs into one | |
pdf_split |
Extract specific pages | |
pdf_rotate |
Rotate pages by 90° | |
pdf_strip_metadata |
privacy | Remove author/title/producer/dates |
image_compress |
image | Lossy re-encode at a target quality |
image_convert |
image | jpeg ↔ png ↔ webp ↔ avif |
image_resize |
image | Resize (never upscales) |
image_strip_metadata |
privacy | Strip EXIF/GPS/ICC/XMP |
image_remove_background |
image | Remove background → transparent PNG (local AI model, no upload) |
image_watermark |
image | Tile a text watermark |
qr_generate |
generate | QR code PNG from text/URL |
Pricing (planned)
| Plan | Price | Quota | Retention |
|---|---|---|---|
| Free | $0 | 100 ops/mo | in-memory |
| Pro | ~$19/mo | 10,000 ops/mo | in-memory |
| Business | ~$99/mo | unlimited | zero-retention guarantee + audit log + BAA |
Run locally
npm install
npm run test:engine # 10 tools, end-to-end, no browser
npm run mcp # stdio MCP server
npm run rest # REST API on :8787 (try key "demo")
Privacy
No file is written to disk by the REST API; the MCP server writes output only to the path you pass. No file contents are logged. The Business tier adds a contractual zero-retention guarantee and audit logging.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。
mcp-server-qdrant
这个仓库展示了如何为向量搜索引擎 Qdrant 创建一个 MCP (Managed Control Plane) 服务器的示例。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。