We have recently been working on an MCP server for the B&R Help content.
The idea is to bring all the B&R specific knowhow from the AS Help into the GitHub Copilot agents (Or any other AI tool).
The server can be hosted on your own PC in either a docker container or via UV.
The local help file will be indexed and loaded into memory for fast and efficient searches and agent grounding.
The MCP Server offers tools for searching and ranking results and gives direct links to your own help instance as well as the online help.
It also provides prompts for fast, focused or in depth searching when needed.
The MCP has support for AS4 and AS6 and can handle multiple AS Help instances simultaneously.
Hi @poulsena , maybe stupid question from me, but I have no deep knowledge about that. We have online help, which is not, to the best of my knowledge, accessible for AI Agent (copilot,etc). Is that the reason why you have started with MCP server? or you need it anyway for functionality (not only help content) for your GitHub Copilot. ??
Hello @kovarj, Yes the Online Help is not indexed by google and therefor is not searchable by an AI Agent. I think you are being rate limited if you try
The MCP will do just that but on your local help - Index everything (on the first start) and present tools that can do iterative AI Assisted search directly on the memory loaded index.
Then the AI agent can automatically look up context when writing code, researching topics or answering support questions. Because the Online Help uses the same URL pattern as the local help package htm /html file locations, the MCP can generate correct links to the Online Help for references.
This also means that the as-help-mcp only presents you (or the AI) with the help content that is installed on your local PC. That is both a benefit and a downside. The downside is that you typically don’t have everything installed, and the benefit is that the AI will never propose libraries or hardware modules that you don’t have available for a given project or setup.
So, if I understand you properly, if online help were indexed by Google, you could use it as well for your GitHub Copilot.
But for you, anyway, it’s better to have a local version and use an MCP server because you can then better control what information is provided to your AI GitHub Copilot.
I’m asking because the community is indexed, so you can use an AI client to access it, but online help is not, and I’m thinking/looking for a possible solution for that.
It would not be as good as having a dedicated MCP for it.
If using a general tool for searching and scraping web content (like: firecrawl, perplexity, apify), it would take more time and fill up the LLM context window with more garbage as well. That is also why there are dedicated MCP servers for aimed content like: GitHub, Context7, Microsoft Learn etc.
We just made one for AS-Help, as we see the need for grounding the AI with the entire help system when working with B&R
