feat(dbt_cli): --target argument support and get_profiles tool

[hanslemm]

This PR includes two main features:

• Support to --target flag in dbt cli commands.
• get_profiles tool, which parses the profiles.yml of the project and returns the available profiles + targets, including their target types (postgres, redshift, bigquery, etc.) and target database names.

Tested the tool call with the prompt: "Can you get the profiles of my project via dbt-mcp get_profiles tool?"
Also did requests with different targets and ran successfully.

How to test this MCP live (example for VSCode mcp.json file)

{
    "inputs": [],
    "servers": {
        "dbt": {
            "command": "uvx",
            "args": [
                "--from",
                "git+https://github.com/hanslemm/dbt-mcp@patch-1",
                "dbt-mcp"
            ],
            "env": {
                "DBT_PATH": <fill>,
                "DBT_PROJECT_DIR": <fill>,
                "DISABLE_SEMANTIC_LAYER": <fill>,
                "DISABLE_DISCOVERY": <fill>,
                "DISABLE_REMOTE": <fill>
            }
        }
    }
}

[DevonFulcher]

Looks great! I just had a few minor comments.

[DevonFulcher]

@b-per should be able to help get the updated diagram uploaded so it appears in the README and pypi.

[hanslemm]

Thanks @DevonFulcher for the feedback.

I've addressed the comments, please check latest changes.

[b-per]

Hi @hanslemm !

I have been testing the PR, and it technically works 🚀 , but I am wondering what exactly you are trying to achieve/solve with it (and/or if we should slightly update the tool).

With this PR, we can

• get the different profiles
• for each profile get the targets
• specify a target for the current project

What I think is missing though is that today we can't know what is the profile used for the current dbt project. So in my case, the LLM tells me that I have 10+ profiles and several targets for each, but most of those targets won't work for my current profile.

I can think of 2 ways to solve this:

• update get_profiles to get_profiles_and_targets and allow the LLM to scope down the targets to the current profile only
• or create a new tool called get_targets that would get the target for the current profile only

What do you think?

[hanslemm]

Good point, since I only have one profile in my case, I didn't think about multiple profiles + targets.

The main point of this PR is allowing the MCP to run on different targets. It is nice to have the chance to tell the LLM which target it should use when reaching the MCP to run the dbt related commands.

[b-per]

Thanks!

In that case, is it OK to rename the tool to get_targets, as this reflects more what you are after, and update the logic it so that it gets the targets of the current profile?

And if it sounds good to you. Let me know if you want to take care of it, otherwise, I can give it a go.