You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Enable the platform broker and Team Broker by default #1010
The chart defaults forge.broker.enabled and forge.broker.teamBroker.enabled to false, while Docker Compose ships the broker on. A default Kubernetes install therefore has no platform broker.
The broker is the prerequisite for MCP tooling and third-party agent access, which route over MQTT and need no Expert token. It is separately one step of Expert configuration, which additionally needs the Expert token. So on Kubernetes both paths are off until the operator edits values.yaml.
Enable both switches by default, or record here why we cannot and what has to change first.
Acceptance:
The original reasons for the off-by-default are written down, with which of them still hold.
If enabled: helm install with no extra flags yields a working platform broker plus Team Broker on a cluster meeting the documented prerequisites, and does not hard-fail on one that does not.
Kubernetes install docs and the self-hosted enablement guide match the resulting default.
it was disabled because a broker was not needed for core functionality
Since the introduction of expert over mqtt and now mcp tooling, this can be considered as needed for core functionality.
enabled by default means users must have the correct version of the emqx operator installed before installing (it is now listed as a pre-req)
customers upgrading will need to ensure all the current pre-reqs are installed before they can upgrade
Can we have gracious errors/fallback in any way, so the configuration can just stand regardless of the emqx operator being installed?
@ppawlowski SH users that have never used FlowFuse Expert, should have no friction gates towards using the FlowFuse MCP tooling we release in 3.0. This configuration requirement creates friction, if its not necessary, please remove it.
If we decide to switch the default forge.broker.enabled value to true, this will create friction for everyone who wants to install FlowFuse Platform on a kubernetes cluster.
Remembering the recent controversies around EMQX, I don't think this is a good idea.
Pulling the key points out of today's Slack thread (link) so they live here rather than only in Slack.
Tokens are not part of this ask.forge.assistant.service.token and forge.expert.service.token are only needed for Expert. The MCP tooling and third-party agent access route over the broker and work without them. So "enabled by default" here means the broker switches only, no per-deployment token values.
The headline is third-party agent access for self-hosted. Assume a self-hosted install that has never set up Expert, so has no tokens. It should still be able to connect an external agent to FlowFuse through the MCP tooling we ship in 3.0, without custom configuration first. The same bridge is needed for Expert too, but Expert then needs the tokens on top.
The gap is Kubernetes only. Docker ships the broker on and needs no custom config. Kubernetes still does. The ask is to align Kubernetes with the way Docker does it.
This is not a blocker, it removes a friction point for customers who want to use the new MCP functionality without having set up Expert first. Ideally resolved before Thursday.
Activity
dimitrieh commentedon Aug 21, 2026
cc: @knolleary @hardillb
knolleary commentedon Aug 21, 2026
To move this forward, it would helpful for us to document:
If there are technical blockers, lets get them written down so we can work out a way forward.
@hardillb @ppawlowski you both have the context here, so your input is needed.
hardillb commentedon Aug 21, 2026
dimitrieh commentedon Aug 21, 2026
Since the introduction of expert over mqtt and now mcp tooling, this can be considered as needed for core functionality.
Can we have gracious errors/fallback in any way, so the configuration can just stand regardless of the emqx operator being installed?
knolleary commentedon Aug 21, 2026
Forgive the basic question, but does that mean:
hardillb commentedon Aug 21, 2026
helm/helm/flowfuse/templates/emqx.yaml
Lines 1 to 4 in 412e834
We may want to update the helm chart to explicitly point to the section of the docs about what version to install
ppawlowski commentedon Aug 21, 2026
I am a little bit lost in the whole conversation - @dimitrieh what kind of problem are we trying to solve here?
dimitrieh commentedon Aug 21, 2026
@ppawlowski SH users that have never used FlowFuse Expert, should have no friction gates towards using the FlowFuse MCP tooling we release in 3.0. This configuration requirement creates friction, if its not necessary, please remove it.
ppawlowski commentedon Aug 21, 2026
I do not see a possibility to enable this by default, mainly because:
truerequires the EMQX operator to be installed firstforge.assistant.service.tokenandforge.expert.service.tokenneed to be provided since these are custom values per deploymentppawlowski commentedon Aug 21, 2026
If we decide to switch the default
forge.broker.enabledvalue totrue, this will create friction for everyone who wants to install FlowFuse Platform on a kubernetes cluster.Remembering the recent controversies around EMQX, I don't think this is a good idea.
knolleary commentedon Aug 21, 2026
@ppawlowski
Lets separate out the EMQX issue from the technical aspects of the 'enable by default' question
That should address your first concern right - we can document the requirement and the helm chart will safe guard against it being missing.
For the tokens - that needs input from other colleagues in terms of how that's intended to work in an 'enabled by default' world.
ppawlowski commentedon Aug 21, 2026
As Ben mentioned:
forge.broker.enabledvalue set totrue, Helm chart installation will hard fail with the message:dimitrieh commentedon Aug 24, 2026
@Steve-Mcl can you take a look here when you can?
cc: @knolleary
dimitrieh commentedon Aug 24, 2026
Pulling the key points out of today's Slack thread (link) so they live here rather than only in Slack.
Tokens are not part of this ask.
forge.assistant.service.tokenandforge.expert.service.tokenare only needed for Expert. The MCP tooling and third-party agent access route over the broker and work without them. So "enabled by default" here means the broker switches only, no per-deployment token values.The headline is third-party agent access for self-hosted. Assume a self-hosted install that has never set up Expert, so has no tokens. It should still be able to connect an external agent to FlowFuse through the MCP tooling we ship in 3.0, without custom configuration first. The same bridge is needed for Expert too, but Expert then needs the tokens on top.
The gap is Kubernetes only. Docker ships the broker on and needs no custom config. Kubernetes still does. The ask is to align Kubernetes with the way Docker does it.
This is not a blocker, it removes a friction point for customers who want to use the new MCP functionality without having set up Expert first. Ideally resolved before Thursday.