[New Feature]: MCP Tool Server for CLI

[jl-0]

Checked for duplicates

Yes - I've already checked

Alternatives considered

Yes - and alternatives don't suffice

Related problems

No response

Describe the feature request

It would be great to develop a MCP server for the SLIM CLI. I think to start a tool server is probably the most usable since it could be incorporated to ADK and Langgraph, but down the road a prompt server would be really interesting as well.

[riverma]

Thanks for this idea @jl-0 - completely agree. @galenatjpl also mentioned this would be useful. It's the to-do!

[riverma]

Some specification suggestions for this:

• At a very basic level, the MCP server should ensure that all available functionality of the slim CLI tool, including the apply, deploy, apply deploy functionality, as well as the list command are all made available for use using natural language from a AI client.
• The MCP server should treat currently existing slim registry assets as MCP resources that can be made available to the AI client and host through the MCP protocol. For example a given README template that is defined in the slim registry as an asset should be made available as an MCP resource so that an AI agent can take that template and fill it out using provided prompts (see below) and the coat base being evaluated.
• The MCP server should define prompts, in fact leverage prompts from pre-existing prompts available for the slim CLI, for each best practice such that the AI host and client can follow precise directions about what work should be done to apply slim best practices to resources that are made available as well as the code that is being reviewed.

Here are some examples of natural language interaction with the slim CI that should be made possible through the creation of an mcp server:

• "List all available slim best practices" -> slim list
• "Apply the slim readme best practice to the current repository without AI" -> slim apply --best-practice-ids readme --repo-dir .
• "Apply the slim readme best practice to the current repository" -> slim apply --best-practice-ids readme --repo-dir . --use-ai MCP
• "Apply the slim readme best practice to the following repo URLs" slim apply --best-practice-ids readme --repo-urls <YOUR_GITHUB_REPO_URL> --use-ai MCP
• "Apply the slim secrets detection best practice assets to the current repository" -> slim apply --best-practice-ids secrets-github --best-practice-ids secrets-precommit --repo-dir . --use-ai MCP
• "Create a documentation website using slim docs website best practice for the following repositories and the output being sent to the following output directory" -> slim apply --best-practice-ids docs-website --repo-urls <YOUR_GITHUB_REPO_URL> --output-dir /path/to/new/docs-website-folder-to-create1 --using-ai MCP

Note that the above needs to support both non-AI mode (which just uses slim best practice templates) as well as AI mode. Non-AI mode needs to be called out explicitly by the term "without AI" or "not with AI" etc. AI mode however should be the default as users will expect so. The AI mode will need to be factored in the slim CLI to support the MCP protocol using a new value of MCP for the pre-existing argument --use-ai. Support for this will be need to be added to the slim CLI. Specifically and currently, the SLIM CLI --use-ai is enacted through the use of the LiteLLM library and AI prompt completion requests are sent directly to LiteLLM to be invoked on specified files depending on the model name that is passed through the command line. For the MCP server however we want to be a little bit smarter than this. Specifically we want to leverage the intelligence of the MCP host, a.k.a. claude code or cursor, to pass in information about files to edit as well as to do the AI editing of files itself within the context of the host. This is obviously a bit more complicated and will not involve the use of LiteLLM per se. However it will be a more powerful model because we won't need to include special code in SLIM CLI to actually perform the AI related updates to the SLIM best practice templates using the prompts. Effectively the MCP server wrapper for SLIM CI allows the SLIM CLI to just be a conduit for bringing together predefined prompts, resources, and target repositories.

In summary, here's the responsibility of the SLIM CLI and of the MCP client:

• SLIM CLI we will be responsible for providing precise prompts, resources that include slim registry assets and templates, and the logic for the underlying basic commands such as list, apply, deploy and apply–deploy
• The MCP client will be responsible for taking the resources mentioned above, such as prompts, and resources and making modifications to the current directory that the MCP client is already looking at as well as providing the necessary contextual information about the directories files. After the modifications have been made, the mcp client should then invoke the respective slim apply, deploy, apply–deploy command needed. This could be a little bit complicated to implement and if it becomes too complicated then we should consider enhancing the SLIM CLI itself to have better contextual capabilities using a third-party library such as LangChain.