Summary
An agent can put a file in your bucket and then work on it there.
What it takes off you is the round trip: the file goes up once, and the cutout, the upscale, the watermark, the PDF conversion and the similarity search all run against the object in place rather than on your machine.
What it is
Tencent's own MCP server for Cloud Object Storage. It reads and writes one bucket, and exposes the data-processing services that run against objects already in it.
What you get
- Uploads in whatever form you have the file: a local path, a string, base64, a buffer, or a URL the server fetches itself
- A bucket listing by prefix, a direct download, or a signed link to one object
- Image work in place: metadata, quality assessment, upscaling, subject cutout, QR-code reading and a text watermark
- Dataset search for images similar to one you supply, or matching a text description
- Asynchronous jobs — a smart video cover and document-to-PDF conversion — each with a second call that collects the result by job id
- A call that reads back the region, bucket and dataset the server was started with
Requirements
Npx on your PATH, plus a region, a bucket, a secret id and key, and a dataset name. All five are passed on the launch line and the server holds them for every call, so one running server serves one bucket.
Setup effort
One command — npx cos-mcp --Region=yourRegion --Bucket=yourBucket --SecretId=yourSecretId --SecretKey=yourSecretKey --DatasetName=yourDatasetname