Summary
Object storage the assistant can read from, not just point at.
The surface is small on purpose: list, read, upload. What makes it worth wiring in is that the read returns the object's contents rather than a link, so a log or a CSV sitting in a bucket becomes something the model can work with in the same turn.
What it is
Qiniu's MCP server for Kodo object storage, built on the S3 protocol, letting an AI client list buckets and objects, read object contents and upload files.
What you get
- Bucket listing across the buckets you configure
- Object listing inside a bucket, by prefix, and continuing where the previous batch stopped
- Object contents read straight into the conversation
- Uploads, either of a local file or of content you supply
- Resource listing that returns 20 items by default, with the tools available to walk the rest
Requirements
Python 3.12 or newer and the uv package manager. In the client config: a Qiniu access key and secret key, the S3 region name and endpoint URL, and the buckets to expose, all in one region and no more than about 20.
Setup effort
One command plus a key — uvx qiniu-mcp-server, then supply credentials