RELEASE 2026: Use B&R MCP Server with Any AI Assistant Using JWT from AS Code 6.7 (GitHub Copilot, Continue Dev & more)

Share Info & Ideas Code & Libraries
,
1

AS Code 6.7 introduces JWT token authentication, allowing you to connect the B&R MCP Server to any AI assistant that supports the MCP — not just AS Copilot.

What is the B&R MCP Server?

The B&R MCP Server is an MCP endpoint that gives your AI assistant access to B&R automation knowledge and community data

Step 1: Get Your JWT Token

After installing AS Code 6.7, your personal JWT token is available at:

File → Preferences → Settings → Extensions → AS Copilot → Ascode-access-token

image
image966×327 21.1 KB

Copy the token value — you’ll need it in the next step.

Step 2: Connect to the B&R MCP Server

The connection setup follows the standard MCP pattern — provide a URL and an authorization header. This works with any MCP-compatible client.

GitHub Copilot Setup

Option A: Command Palette

  1. Open the Command Palette (Ctrl+Shift+P)
  2. Run “MCP: Add Server…”
  3. Choose HTTP as the transport type
  4. Paste the URL: https://app-prod-bnr-as-copilot-api.azurewebsites.net/mcp/
  5. Enter a name, e.g. B&R MCP Server

This command scaffolds the configuration for you and opens the config file where you can add the authorization header.

Option B: Edit mcp.json directly

GitHub Copilot stores its MCP server configurations in a file called mcp.json. You can find it at:

  • User-level (applies to all workspaces): %APPDATA%\Code\User\mcp.json on Windows, ~/.config/Code/User/mcp.json on Linux/macOS
  • Workspace-level (applies to one project only): .vscode/mcp.json in your project root

Add the following entry inside the "servers" object:

"B&R MCP Server": {
        "type": "http",
        "url": "https://app-prod-bnr-as-copilot-api.azurewebsites.net/mcp/",
        "headers": {
            "Authorization": "Bearer <YOUR_JWT_TOKEN>"
        }
    }
}

After saving, Copilot automatically detects the server. You can verify the connection via the MCP tools icon in the Chat panel.

Other AI Assistants

The setup is the same for any MCP-compatible client — just provide the URL and Bearer token from AS Code in your client’s MCP configuration.

TL;DR: Install AS Code 6.7 → copy JWT from settings → add the B&R MCP Server to your AI assistant’s MCP config → get automation-aware AI responses everywhere.

14 Likes
NEWS: Official B&R MCP sever was released
3

Perfect, this is what we have been looking for for a long time. Definitely, I will test it in the near future. :star_struck:

2 Likes
4

Hi,

I didn’t find the “MCP” command palette in Automation Studio Code 6.7.

Do I need to enable anything else?

Thanks

5

@simone.lonardoni I think MCP pallete command is for VS code

1 Like
6

In Automation Studio Code out of the box you can use ASCopilot with the extension that is already installed. This extensions is using the MCP server mentioned here to give you the answers inside ASCode directly.

AS Code:

image
image499×1109 22.4 KB

The post here mentions how to use the MCP server (which in the end is ASCopilot revamped) inside your own toolings.
So if you have for example a GitHub Copilot, Claude Code, etc subscription, you can now use ASCopilot as an MCP server directly in that harness.

VS Code:
Ctrl+Shift+P

image
image640×112 4.51 KB

image
image656×107 7.34 KB

image

image
image614×113 6.06 KB

image
image908×154 6.75 KB

AS Code:

image
image1124×344 28.2 KB

5 Likes
7

@tomas.miculka, @phaehnel

Thanks

2 Likes
8

B&R automation knowledge and community data → what is it exactly?

1 Like
9

Afaik it includes AS Online Help, Community posts and AS6 sample projects for each technology.

3 Likes
10

Is it possible to include a local folder with multiple files in AS Code Agent?

11

What really is new in this MCP server that we didn’t have earlier?

Can AI compile the project?

Can AI transfer the project?

Can it run simulation?

Can it read/write variables on running target / simulator?

12

Hi @bartosz.pawlik,

I’m absolutely no AI expert, nevertheless I try to answer what I know about (or at least what I think I’ve understand about the topic :wink: ).
I’m not sure if your question “… that we didn’t have earlier” is related to “what every AS Code user has had earlier”, or to “what a B&R employee has had earlier” (because internally, a B&R employee have access to some other tools then all of the users have (yet))?

From my understanding, the main difference from “AS Code user” point of view of “before” and “now” is:

  • before, (just) AS Code was able to access one specific LLM that was trained with B&R specific data
  • now, any LLM you want to use can access B&R specific data by the MCP server - so now you have the freedom of choice a) which LLM you want to use, and b) which environment you want to use for that access (AS Code, VS Code, others … depending on your use-case).

This change makes it much easier or even possible to use different LLMs for different tasks, using Agents, participate to 100% of the evolution of LLMs, update the B&R specific database much faster (and generally enrich the specific database with other, maybe company- or use-case specific data by providing that data via other MCP servers), and more.

About your questions if AI can compile, transfer, run, read/write variables, and so on:

  • in general: yes
  • out-of-the-box delivered by B&R: no (not yet)

By now, the MCP server from above delivers B&R specific data to a common LLM.
That means if you’re already using any agentic framework in Windows that for example has skills that are able to execute shell commands (which are often also commanded by MCP servers), then the AI agent framework for example uses a skill that is able to start a build process by using the Automation Studio command line interface BR.AS.Build.exe, and the MCP server mentioned in this thread would deliver the information to the LLM how this interface has to be used …
… to all AI experts out there - I know this example is a bit shallow and incomplete , but it should just explain a little bit the relationship between LLM, MCP, Agent and Skill.

Why “out-of-the-box delivered by B&R: no (not yet)”?
Afaik, internally there’s work ongoing into the direction to make also “agent-friendly” tools / skills (I think accessible via MCP servers?), custom prompts / instructions, and so on available to realize this parts as comfortable as possible. But I’m not aware about the “when” and “how”.

Best regards!

PS: after writing this post, I asked AI to explain the domains LLM, Agent, Skill, MCP in “easy words”. This was the answer - I personally like it and therefore share it here, too:

:puzzle_piece: The Four Domains Defined

1. LLM (Large Language Model)

  • What it is: A smart text calculator.
  • Its role: It predicts the next best words based on patterns.
  • Analogy: A brilliant scholar who has read every book but sits in a dark room with no internet or hands. [1, 2, 3, 4, 5]

2. Agent

  • What it is: An autonomous software system driven by an LLM.
  • Its role: It loops through a cycle of Thinking (using the LLM), Planning, and Acting.
  • Analogy: A smart assistant who takes your goal, breaks it into steps, and works until it is done. [1, 2, 3, 4, 5]

3. Skill

  • What it is: A specific, hardcoded capability.
  • Its role: It allows the system to do one exact task perfectly, like checking the weather or calculating taxes.
  • Analogy: A single app on your smartphone. [1, 2, 3, 4]

4. MCP (Model Context Protocol)

  • What it is: An open standard created by Anthropic for connecting AI to data and tools.
  • Its role: It acts as a universal plug. It lets any LLM safely read databases, view files, or use Skills without custom code for each one.
  • Analogy: A standard USB-C port that lets any device talk to any accessory. [1, 2, 3, 4, 5]

Step-by-Step Workflow Example

Imagine you tell an Agent: “Find my top three clients in my database and email them a thank-you note.”

  1. The Agent Receives the Goal: The Agent takes your prompt and creates a plan.
  2. The LLM Thinks: The Agent asks its LLM “brain” how to start. The LLM says: “First, we must fetch the client data from the database.”
  3. The MCP Connects: The Agent uses the MCP standard to securely open a connection to your private database.
  4. The Skill Executes: Through that MCP connection, the Agent triggers a specific Database Read Skill to pull the client names.
  5. The Loop Repeits: The data goes back to the LLM. The LLM drafts the emails.
  6. The Final Action: The Agent uses an Email Skill via MCP to send the messages. [1, 2, 3, 4, 5]
3 Likes
13

Thanks a lot for your explanation, actually that is exactly what I expected.

We have already seen mcp servers that digested AS Help, so this one is just better at doing B&R stuff. What I’m missing here is:

“out-of-the-box delivered by B&R: no (not yet)”?

This part is crucial. I mean, there are existing solutions to this, like this one: AndrewMusser/br-automation-mcp but:

  1. It’s unofficial
  2. It doesn’t work out of the box (eg. hardcoded timeout or warnings that cause build fails)

I saw LinkedIn posts that led to our website Agentic Bridge with beautiful videos and promises of agentic workflows. Yet, here we are, working directly on files and external executables. Neither AS Code nor AS6 changed anything in this regard.

Citation from our website:

Automation Studio Agentic Bridge expands AI beyond code generation, enabling intelligent agents to execute engineering workflows, validate results, and actively participate in real projects.

Can someone explain to me how it can validate results?

The way I see it, AS Agentic Bridge is unable to do it

14

Hi,

I fully agree, and I’m pretty sure the project accountables do too!

In my interpretation, the Agentic Bridge is an announcement of a upcoming “product”/“framework”/“tool collection” (or however we’ll call it), where the MCP server is just the first step already available, and more is coming (hopefully soon).

Again, I agree. From my understanding, reading / writing process variables (+ some more things) by an agent through a skill must be available to validate results.

Right now, I think this is only possible by creating an own skill using OPCUA or PVI like for example Andrew did (I haven’t tried, but there some open source wrappers like pvi.py out there, which in theory would make it possible to create an agent without developing it totally from scratch)…
… but it would not be nice if everybody has to do this on his own.
Fortunately, as far as I know, this is also not the plan :slight_smile: → exactly such skills e.g. for reading / writing variables and more are in the pipeline, but need some more time to grow.

If the’re concrete customer requests, I would propose to clarify the possibilities with the project owners via the sales representative → I really don’t know if it’s already possible and what would be prerequisites.
As B&R employees, we can already get access to some of the stuff for testing and contribution (of course in coordination with the line management).

BG Alex

1 Like
15

It is a combination of skills/agent template, CLI and a knowledge MCP server.

So it will be able to do everything you mention there and actually even more.

Get access to B&R domain knowledge, skills for most B&R technologies to get already the basics into your harness of choice, building + testing on a running system (PLC or Simulation) and much more

1 Like
16

The B&R MCP Server is configured as an HTTP endpoint using a JWT token. I would like to better understand the data protection implications.

Could you please clarify:

  • Is the user’s IP address or other connection metadata logged?
  • Are chat requests and search queries stored permanently?
  • How long are requests and logs retained?
  • Are submitted queries or code snippets used to improve AI models?
  • Is any data shared with third parties?
  • Does the MCP server receive only the explicitly submitted prompt and context, or can it access local project files automatically?
  • Are project files, PLC programs, machine IP addresses, or other customer data stored or processed?

I understand that the MCP server cannot access local files or a PLC directly unless the AI client explicitly sends this information. I would appreciate confirmation of this and any links to the relevant privacy policy or data-processing documentation.

17

Excellent question, thank you @c496652

TL;DR:
B&R/ABB has no access to your data and data is not used to train any AI models.

A bit more details:

4 Likes