One of the great things about modern AI editor workflows is how it makes it
easier to get started. Normally when you open a text editor, you have an empty
canvas and don’t know where to start. AI tools let you describe what you want
and help you get started doing it.
“We’ve all been excited about AI editors making development fast and just plain fun.”
-Most developers, probably
Today we’re happy to announce that we’re making it even easier to get started
with Tigris in your AI editor workflow. If you want to get to the part where you
can plug configs into your AI editor and get started, head to Getting Started and get off to vibe coding your next generation B2B SaaS as a service.
Abdullah just started at Tigris a week ago (Welcome!) and has already built
something that will make it easier for you to make object storage a native part
of your development workflow: a Model Context Protocol (MCP) server for Tigris.
This enables you to manage your buckets and objects with plain language in your AI capable editor.
Just say “make me a bucket for this project” and it’ll go do that. Want files in the bucket? Just ask it to upload a file; it’ll make it happen.
The vision
We want your developer experience with Tigris to be as seamless, unsurprising,
and natural as possible. What’s more natural than natural language? Getting this
set up was a breeze. Tigris is compatible with S3, so all Abdullah had to do was
glue S3 calls to the MCP library. Everything was already there, well-tested, and
ready to go. And it’s just object storage: there’s no chance you’ll accidentally
spin up an expensive service and get a surprise bill.
Of note: many other MCP servers will try and do much more than they need to. Our
MCP server just does object storage. There’s no possibility of it spinning up
expensive servers and saddling you with a surprise bill with an unreasonable
number of zeroes in it.
Getting started
To get started, create some access keys and then install our MCP server:
Edit your config file
Add this snippet to your claude_desktop_config.json or mcp.json for Cursor
AI
{
"mcpServers": {
"tigris-mcp-server": {
"command": "npx",
"args": ["-y", "@tigrisdata/tigris-mcp-server", "run"],
"env": {
"AWS_ACCESS_KEY_ID": "YOUR_AWS_ACCESS_KEY_ID",
"AWS_SECRET_ACCESS_KEY": "YOUR_AWS_SECRET_ACCESS_KEY",
"AWS_ENDPOINT_URL_S3": "https://fly.storage.tigris.dev"
}
}
}
}
Run the init script
Run the init script in your terminal:
npx -y @tigrisdata/tigris-mcp-server init
Then ask your editor to make you a bucket for a project and it will! More instructions are on the official npm package here.
Trust
AI editors and tooling are really cool, but there’s some key things you should
be aware of before you blindly trust this. The Model Context Protocol ecosystem
is still very new, so there will almost certainly be problems we solve together
over time. There are also inherent risks involved in giving any tool access to
your cloud storage accounts or filesystem.
The Model Context Protocol server will run in the same level of sandboxing as your editor does. Be careful with what you install and always double-check what you run before you run it.
In order to make this as safe as possible, we’ve made the Model Context Protocol
also available as a docker container. This means you can run it in a sandboxed
environment and not have to worry about it having access to your local
filesystem. You can run it in a container and container have access to a
specific directory on your local filesystem. This is a great way to make sure
that the Model Context Protocol server can only access the files you want it to.
Edit your config file
Add this snippet to your claude_desktop_config.json or mcp.json for Cursor
AI. Please note that CURRENT_USER references the user running the command.
{
"mcpServers": {
"tigris-mcp-server": {
"command": "docker",
"args": [
"run",
"-e",
"AWS_ACCESS_KEY_ID",
"-e",
"AWS_SECRET_ACCESS_KEY",
"-e",
"AWS_ENDPOINT_URL_S3",
"--network",
"host",
"--name",
"tigris-mcp-server-claude-for-desktop", // tigris-mcp-server-cursor for Cursor AI
"-i",
"-v",
"tigris-mcp-server:/app/dist",
"--rm",
"--mount",
"type=bind,src=/Users/CURRENT_USER/tigris-mcp-server,dst=/Users/CURRENT_USER/tigris-mcp-server",
"quay.io/tigrisdata/tigris-mcp-server:latest"
],
"env": {
"AWS_ACCESS_KEY_ID": "YOUR_AWS_ACCESS_KEY_ID",
"AWS_SECRET_ACCESS_KEY": "YOUR_AWS_SECRET_ACCESS_KEY",
"AWS_ENDPOINT_URL_S3": "https://fly.storage.tigris.dev"
}
}
}
}
Run the init script
Run the init script in your terminal and select Docker as option when
prompted
npx -y @tigrisdata/tigris-mcp-server init
This Model Context Protocol server will run with the full power and authority of any credentials you give it. Be very careful about typos with object names that have similar token distances.
Additionally, AI tools are fundamentally built around random behavior and will
have unexpected results at times. Sometimes it takes the AI a couple tries to
learn what you want to do. Be very careful, as typos in an AI context can have
much more drastic consequences than they can in normal contexts. We don’t want
you to lose data you need. For example, when you run the DeleteBucket call,
you are not allowed to do this unless the bucket has no data in it.
In order to be as transparent as possible, we’ve made our Model Context Protocol server open source and are actively monitoring that repository.
We’re making this as safe and reliable as possible. Part of this is the scope
reduction we mentioned earlier: we’re only managing your buckets and objects. The other part is by going out of our way to make this tool as boring as possible. Boring code is easy to understand, easy to maintain, and easy to learn from. We hope that this will help you make the exciting parts of your program while leaving the boilerplate to machines.
We hope this will make using Tigris absolutely frictionless and that you can
learn how S3’s API works in the process. Not to mention, we want you to get out
there and build things!
This article was originally published on April 3, 2025 at tigrisdata.com/blog
Top comments (0)