[FEATURE]: Add Dockerfile to Unity Catalog MCP Server

[ognis1205]

Is there an existing issue for this?

• I have searched the existing issues

Problem statement

Currently, the Unity Catalog MCP Server supports local development via uv, but it lacks a Dockerfile for containerized execution.

Proposed Solution

• Add a well-defined Dockerfile that builds the Unity Catalog MCP Server

Additional Context

N/A

[ognis1205]

Can I take this one? Does this make sense?

[smurching]

@ognis1205 thanks for filing! Yes this makes sense, happy to review a PR for this :)

[ognis1205]

take

[JordiNeil]

Hi, current structure have a single pyproject.toml file, for both MCPs, UC and developer tools. How would we handle two servers running in the same docker container? Or should we split both MCPs where each one has its own pyproject.toml, dependencies and dockerfile?

[ognis1205]

@JordiNeil

I vote in favor of preparing a separate pyproject.yaml and Dockerfile for each MCP server implementation. For example, I don't think the developer-tools server needs to depend on unitycatalog-ai (though I'm not fully familiar with the details). While it's also an option to handle this as a workspace, similar to Rust's cargo, I wouldn't recommend it given the current state of uv's implementation:

https://docs.astral.sh/uv/concepts/projects/workspaces/

[JordiNeil]

I agree with @ognis1205

To have a single pyproject we would need to create a common entrypoint for both MCP servers and also a single docker.
I think in that case it doesn't makes sense to have two MCP servers packaged together.

[ognis1205]

@JordiNeil @smurching

It might be a good idea to file a GitHub Discussion or Issue to discuss this.

[JordiNeil]

@renardeinside how would it work using Dockerfile and a single pyproject? I'm not so familiar with uv.

If we have a single pyproject with two scripts, one for UC and another for developer tools, should we parametrize the dockerfile to know when to run one or the other?

Kind of
docker run --server=uc-mcp

vs

docker run --server=dev-tools

Do you have any other idea?

[renardeinside]

Hey @JordiNeil ,

yes, I think this would be the approach. It can work like this:

1. single pyproject.toml with dependency groups unity-catalog and dev-tools

From here we have options:

1. Option one - lightweight containers, separate tag per server:

For example:

docker pull databrickslabs/mcp:<version>-unitycatalog
docker pull databrickslabs/mcp:<version>-devtools

This approach is followed in Python dockerfile - take a look here for reference.

2. Option 2 - single fat container:

docker pull databrickslabs/mcp:<version>-full

And then:

docker run databrickslabs/mcp:<version>-full mcp-server

I think they can coexist, just we'll need to make necessary adjustments in CD process.