Add an LLM-assisted interface for workflows, entity definitions, and API integration

[fibenacci]

Goal

Build an MCP server for RDataCore instead of adding LLM-specific logic into the application itself.

The server should connect to the existing RDataCore API via api_key and use the current permission model as-is.

Why

• no separate AI logic inside RDataCore
• no duplicated business logic
• the API stays the single integration surface
• desktop and IDE LLM clients can use RDataCore through a standard MCP setup

Target users

• developers who want less custom integration work so non-technical staff can operate data workflows through LLM clients
• technical or AI-driven management users who can install and use MCP in desktop or IDE tools

Scope

• add a small Rust crate that runs as an MCP server
• connect it to RDataCore only through the existing API
• authenticate with api_key
• expose a first set of useful MCP tools for RDataCore operations
• build binaries for macOS, Linux, and Windows
• distribute it through an npm package in the usual MCP style

Non-goals

• no embedded LLM assistant in r_data_core
• no permission bypass
• no separate domain logic outside the existing API

First capabilities

• inspect entity definitions
• read and search entities
• inspect workflows
• expose useful system or capability metadata

Write operations can follow later once the read-oriented path is stable.

Success criteria

• RDataCore remains free of separate LLM orchestration logic
• an MCP-compatible desktop or IDE client can connect to RDataCore
• permissions are enforced through the existing api_key
• installation is simple on major operating systems

[BentBr]

Pleaes refactor this issue:

We don't want to have separate logic in RDataCore. We should implement an MCP server which is connecting to the API as per given permission (via api_key). Therefore we should write a little crate which compiles to a binary for all major os and bundle it via some npm package. the usual MCP logic.

Our target group is devs which want to reduce complex development on their own to enable (mostly) non-technical personal to cook their data and on the other hand management which is (mostly) technical and AI driven which should be capable of installing and using an MCP.
Our target groups should all use the desktop or IDE versions of the major LLMs

p.s. don use caps in titles. this is too much bullshit bingo for me.
use lables

[fibenacci]

@BentBr Done. I refactiored the desciption of this issue and removed the caps in the title.