ISSUE-10 Initial implementation for MCP Start Events. Includes a plug… by pieter-schutte · Pull Request #14 · finos/fluxnova-ai

pieter-schutte · 2026-02-11T10:14:26Z

ISSUE-10 Initial implementation for MCP Start Events. Includes a plugin component that will render a new pallete option called MCP Start Event and a corresponding java components that will register MCP components for each MCP Start event, on Startup and process deployment.

NB:

This code still need hardening, and decision needs to be taken about security. My current thought is to leave it to the Spring container layer.
For now I have kept both the plugin and the corresponding java code within this repo. There was a proposal to add the plugin in fluxnova-bpm-platform but I feel that will result in bad user experience as the Java dependency will not be included.
I am using npm module camunda-modeler-plugin-helpers. Do we need to fork this module as well?
I have never written a Camunda / Fluxnova plugin, very open to feedback and changes.
@EdsGianesi - stylesheets and icons are just ai generated, can you please take a look.

…in component that will render a new pallete option called MCP Start Event and a corresponding java components that will register MCP components for each MCP Start event, on Startup and process deployment.

EdsGianesi

From the UX viewpoint, it looks appropriate. As. our AI offerings grow, We may need to review these early elements.

pieter-schutte · 2026-02-12T08:52:31Z

I've had one error report from Olivier so far. Please report any issues here.

I've tested on 1.1.1-dev

pieter-schutte · 2026-02-12T14:57:45Z

plugins/mcp-tool-plugin/client/client.js

@@ -0,0 +1,58 @@
+'use strict';
+
+var registerBpmnJSPlugin = require('camunda-modeler-plugin-helpers').registerBpmnJSPlugin;


We have been discussing this internally. The repo is under MIT: https://github.com/camunda/camunda-modeler-plugin-helpers

Do we need to fork this? Would like to get a view from our FINOS experts on using it. @opoupeney

Alternatively, Fidelity seems to have a lot of experience in plugins, perhaps advise a different approach @HarishMalavade

We will able to add this plugin to modeler release whenever we decide to make this available as part of Fluxnova, I was not able to build this plugin in local, I think one of npm library asked for funding. @pieter-schutte is it possible for you to upload compiled plugin so we can push the generated nodeJS artifacts to plugin folder for the local testing, I think the readme was missing where and how to copy plugin post build was success

@swalsh01 can you have someone from our Modeler team look at mcp-tool-plugin folder and evaluate what it takes to make it as part of Fluxnova modeler repo?

I think one of npm library asked for funding

That shouldn't block you from installing it (those notices are informational only I believe). There's probably another issue blocking build. If you post more details of any errors on build that might help with debug.

Thanks Chris. It was my local thing, I was able to resolve the issue. For now I just copied the compiled node modules into local plugin folder. may be As I have called out Sean and team can look at making it part of our regular plugin contribution

HarishMalavade

Couple of general Cosmetic comments

Should we start creating general Plugins or extensions repo and put all MCP Backend code into that folder and plugin will go into modeler repo
Can we have better logos for some of these plugins where possible
How can we point developers to MCP related documentation on how to set up MCP for those who are new to it.

PieterSchutte · 2026-02-13T20:24:43Z

I have created a demo which contains all the MCP plumbing you need, plus two examples. I have set the repo to private for now. Fluxnova Maintainers, please ping me via github for link and access.

kziemski · 2026-02-17T16:42:07Z

@pieter-schutte hey happy to take a look

prajwolbhandari1 · 2026-02-17T18:22:57Z

Maintainers

@pieter-schutte would love to take a look at it. Let me know the best way to look at that example. Thx.

HarishMalavade · 2026-02-17T22:53:54Z

plugins/README.md

@@ -0,0 +1,230 @@
+# MCP Tool Plugin


Few more thought process after trying to run this code in local.

It would be good to call out prerequsites for running this plugin as part of Fluxnova Engine - Going through code today, I understood this is a plugin that needs to be registered and added to Fluxnova Engine

It would be good if we can modify this repo as an full fledged springboot application or an example that we can simply start, For now like you highlighted we keep the plugin code here and discuss on how we can make some of these plugins available as extensions to be embeded into Clients application on demand or decide to make this part of default Fluxnova-bpm-platform

Can we also update instructions on how get it running locally - especially differentiating the MCP Host and MCP Server pieces - My Interpretation correct me if I am wrong is this plugin is how we can have BPM specific MCP server allowing us to start any processes via MCP client (Which can be in host application like chat bot) and MCP host is the actual model we design and deploy. Is it fair to assume that?

pieter-schutte · 2026-02-18T12:58:36Z

I will resubmit the example project to the fluxnova-examples repository, but it will take a day or two.

@HarishMalavade As I cant upload the original PPT due to restrictions. here is the second best option, which will also be added to the various README's:

Overview of where it fits:

flowchart LR
    AI[🤖 AI]

subgraph Potential["Potential Architecture"]
subgraph Existing["Existing AI Application"]
subgraph AI_Impl["AI Implementation"]
subgraph AI_App["AI Application"]
C1[MCP Client 1]
C2[MCP Client 2]
C3[MCP Client 3]
end
end

subgraph Service_Layer["Service Layer"]
subgraph Embedded["Embedded Service #2"]
S2[MCP Server 2]
end

S1[MCP Server 1]

subgraph Tools1["Tools"]
T1["Tool #1-5"]
end
end
end

subgraph Fluxnova["Fluxnova Runtime"]
S3[MCP Server 3]
FD[Fluxnova AI MCP Dependency]
end

subgraph Tools2["Fluxnova Tools"]
T2[BPMN Diagram A]
T3[BPMN Diagram B]
T4[BPMN Diagram C]
end
end

    AI <--> AI_App

    C1 -->|MCP Protocol| S1
    C1 -->|MCP Protocol| S2
    C2 -->|MCP Protocol| S2
    C3 -->|MCP Protocol| S3

    S1 -->|API| Tools1
    S3 --> Tools2

    style Fluxnova fill:#90EE90,stroke:#228B22,color:#000
    style Tools2 fill:#90EE90,stroke:#228B22,color:#000
    style S3 fill:#90EE90,stroke:#228B22,color:#000
    style FD fill:#90EE90,stroke:#228B22,color:#000
    style T2 fill:#90EE90,stroke:#228B22,color:#000
    style T3 fill:#90EE90,stroke:#228B22,color:#000
    style T4 fill:#90EE90,stroke:#228B22,color:#000

MCP Client Config for pre-existing MCP client. Below shows possible Spring-AI implementation:

    mcp:
      client:
        name: mcp-client
        version: 0.1.1
        toolcallback:
          enabled: true
        streamable-http:
          connections:
            mcp-tools-fluxnova:
              url: http://mcp-tools-fluxnova:8080}/mcp

MCP Server Config hosted in Fluxnova Runtime:

    <dependency>
      <!-- This dependency is not yet live and needs to be build from this repo -->
      <groupId>org.finos.fluxnova.ai.mcp</groupId>
      <artifactId>mcp-fluxnova-extension</artifactId>
      <version>0.1.0</version>
    </dependency>

MCP Server Config hosted in Fluxnova Runtime, which is Spring AI specific:

spring.ai.mcp.server.name=my-mcp-server
spring.ai.mcp.server.version=0.0.1
spring.ai.mcp.server.protocol=streamable
spring.ai.mcp.server.stdio=false
spring.ai.mcp.server.type=sync

HarishMalavade · 2026-02-18T17:00:40Z

I will resubmit the example project to the fluxnova-examples repository, but it will take a day or two.

@HarishMalavade As I cant upload the original PPT due to restrictions. here is the second best option, which will also be added to the various README's:

Overview of where it fits:

flowchart LR
    AI[🤖 AI]

subgraph Potential["Potential Architecture"]
subgraph Existing["Existing AI Application"]
subgraph AI_Impl["AI Implementation"]
subgraph AI_App["AI Application"]
C1[MCP Client 1]
C2[MCP Client 2]
C3[MCP Client 3]
end
end

subgraph Service_Layer["Service Layer"]
subgraph Embedded["Embedded Service #2"]
S2[MCP Server 2]
end

S1[MCP Server 1]

subgraph Tools1["Tools"]
T1["Tool #1-5"]
end
end
end

subgraph Fluxnova["Fluxnova Runtime"]
S3[MCP Server 3]
FD[Fluxnova AI MCP Dependency]
end

subgraph Tools2["Fluxnova Tools"]
T2[BPMN Diagram A]
T3[BPMN Diagram B]
T4[BPMN Diagram C]
end
end

    AI <--> AI_App

    C1 -->|MCP Protocol| S1
    C1 -->|MCP Protocol| S2
    C2 -->|MCP Protocol| S2
    C3 -->|MCP Protocol| S3

    S1 -->|API| Tools1
    S3 --> Tools2

    style Fluxnova fill:#90EE90,stroke:#228B22,color:#000
    style Tools2 fill:#90EE90,stroke:#228B22,color:#000
    style S3 fill:#90EE90,stroke:#228B22,color:#000
    style FD fill:#90EE90,stroke:#228B22,color:#000
    style T2 fill:#90EE90,stroke:#228B22,color:#000
    style T3 fill:#90EE90,stroke:#228B22,color:#000
    style T4 fill:#90EE90,stroke:#228B22,color:#000

MCP Client Config for pre-existing MCP client. Below shows possible Spring-AI implementation:

    mcp:
      client:
        name: mcp-client
        version: 0.1.1
        toolcallback:
          enabled: true
        streamable-http:
          connections:
            mcp-tools-fluxnova:
              url: http://mcp-tools-fluxnova:8080}/mcp

MCP Server Config hosted in Fluxnova Runtime:

    <dependency>
      <!-- This dependency is not yet live and needs to be build from this repo -->
      <groupId>org.finos.fluxnova.ai.mcp</groupId>
      <artifactId>mcp-fluxnova-extension</artifactId>
      <version>0.1.0</version>
    </dependency>

MCP Server Config hosted in Fluxnova Runtime, which is Spring AI specific:

spring.ai.mcp.server.name=my-mcp-server
spring.ai.mcp.server.version=0.0.1
spring.ai.mcp.server.protocol=streamable
spring.ai.mcp.server.stdio=false
spring.ai.mcp.server.type=sync

Thank you Pieter, This is helpful. will wait for your example springboot app to be contributed and try further

bryankenneally · 2026-02-19T12:15:41Z

plugins/mcp-tool-plugin/webpack.config.js

We can write the compiled version of the plugin to a "dist" folder and copy over the CSS to same. Removing the compiled version from source control would also eliminate the need to keep the src in sync with the compiled version in source control. We would just have to remember to build the plugin before using it.

const CamundaModelerWebpackPlugin = require("camunda-modeler-webpack-plugin"); const CopyPlugin = require("copy-webpack-plugin"); const { join, resolve } = require("node:path"); const outputDir = resolve(__dirname, "./dist"); module.exports = { mode: "development", entry: "./client/client.js", output: { path: outputDir, filename: "client.js", }, devtool: "cheap-module-source-map", plugins: [ new CamundaModelerWebpackPlugin(), new CopyPlugin({ patterns: [ { from: resolve(__dirname, "./client/styles.css"), to: join(outputDir, "./styles.css"), }, ], }), ], };

bryankenneally · 2026-02-19T12:22:29Z

plugins/mcp-tool-plugin/package.json

+    "domify": "^1.4.0"
+  },
+  "devDependencies": {
+    "camunda-modeler-plugin-helpers": "^3.0.0",


We could use the latest version of "camunda-modeler-plugin-helpers".

Using copy-webpack-plugin we can copy the src CSS into the "dist" folder at build time. camunda-modeler-webpack-plugin can also be used to ensure compatibility with the modeler's plugin system.

"camunda-modeler-plugin-helpers": "^5.1.0", "camunda-modeler-webpack-plugin": "^0.2.0", "copy-webpack-plugin": "^13.0.1",

bryankenneally · 2026-02-19T12:24:09Z

.gitignore

-website/yarn-error.log*
+website/yarn-error.log*
+
+/.idea


Suggest adding "dist/" here if we go ahead with writing compiled versions of plugins to a separate directory

bryankenneally · 2026-02-19T12:25:17Z

plugins/mcp-tool-plugin/index.js

+
+module.exports = {
+  name: 'MCP Tool Plugin',
+  style: './client/styles.css',


Suggest registering the plugin using compiled version from the "dist/" folder

style: "./dist/styles.css", script: "./dist/client.js",

prajwolbhandari1 · 2026-02-24T15:51:55Z

@pieter-schutte any chance that you made your repo public?

pieter-schutte · 2026-02-25T11:05:59Z

@pieter-schutte any chance that you made your repo public?

Please check your github notifications / messages. there should be an invite from about a week ago.

pieter-schutte · 2026-02-25T11:51:29Z

@bryankenneally @HarishMalavade you suggested taking over the plugin component? I would be very happy for you to do that, @bryankenneally seems to know a log more about plugins than myself.

bryankenneally · 2026-02-25T20:48:13Z

@pieter-schutte any chance that you made your repo public?

Please check your github notifications / messages. there should be an invite from about a week ago.

Hey @pieter-schutte, is there any way of making an example available that I could view? I'm not in the maintainers group.

pieter-schutte · 2026-02-26T12:44:02Z

@pieter-schutte any chance that you made your repo public?

Please check your github notifications / messages. there should be an invite from about a week ago.

Hey @pieter-schutte, is there any way of making an example available that I could view? I'm not in the maintainers group.

It might be easier to test in isolation. just enable the mcp settings in an existing fluxnova application:

0d2e1c5#diff-0ffbcc3871158493227e1de4f1875a4d9a5de1b776d58da0a7c7e01ddc59f118R41-R67

Add the following dependencies to your pom.xml:

<dependencies>
    <!-- Spring AI MCP Server -->
    <dependency>
        <groupId>org.springframework.ai</groupId>
        <artifactId>spring-ai-starter-mcp-server-webmvc</artifactId>
    </dependency>

    <!-- MCP Fluxnova Extension -->
    <dependency>
        <groupId>io.mcp.tools</groupId>
        <artifactId>mcp-fluxnova-extension</artifactId>
    </dependency>
</dependencies>

Spring MCP Properties

Spring AI needs to be configured. Below is an example, but please refer to Spring AI documentation.

# MCP Server Configuration
spring.ai.mcp.server.name=fluxnova-mcp-server
spring.ai.mcp.server.version=1.0.0
spring.ai.mcp.server.protocol=streamable
spring.ai.mcp.server.stdio=false
spring.ai.mcp.server.type=sync

The MCP components should be directly testable via npx @modelcontextprotocol/inspector. After connecting to your Fluxnova server, you can invoke the process via the Tools menu.

HarishMalavade · 2026-03-03T14:40:05Z

@pieter-schutte Sharing my thoughts on the plugin approach that we discussed in last week's maintainers call
First of all, I really appreciate and excited for all the great AI contributions Natwest is brining to Fluxnova, this is a great win for Fluxnova community to leverage the opensource collaboration

Advantages of enabling capabilities like AI integrations as plugin

plugins hook into engine bootstrap and configurations allowing organizations to enable/disable on demand basis vs them being part of core engine - This is important because organizations might have different stance and custom integration needed especially for things like AI, MCP, Agentic, Authorization, Events etc
Since plugins are configuration based they can be easily configured with maven dependency and spring config annotation that improves the portability across environments
This allows us cleaner separation of Core engine capabilities which is more around BPMN and DMN engines or any capabilities needed for orchestration and automation with extended capabilities like MCP integration, custom security integration etc
This would also allow us to build granular functionality independently with no disruption to core functionality.

So I recommend that this

MCP backend java code integration become a plugin
Modeler code is added as part of Modeler plugin - which does not impact runtime anyways

pieter-schutte · 2026-03-03T14:46:16Z

@pieter-schutte any chance that you made your repo public?

Please check your github notifications / messages. there should be an invite from about a week ago.

Hey @pieter-schutte, is there any way of making an example available that I could view? I'm not in the maintainers group.

It might be easier to test in isolation. just enable the mcp settings in an existing fluxnova application:

0d2e1c5#diff-0ffbcc3871158493227e1de4f1875a4d9a5de1b776d58da0a7c7e01ddc59f118R41-R67

Add the following dependencies to your pom.xml:
<dependencies>
    
    <dependency>
        <groupId>org.springframework.ai</groupId>
        <artifactId>spring-ai-starter-mcp-server-webmvc</artifactId>
    </dependency>

    
    <dependency>
        <groupId>io.mcp.tools</groupId>
        <artifactId>mcp-fluxnova-extension</artifactId>
    </dependency>
</dependencies>
Spring MCP Properties

Spring AI needs to be configured. Below is an example, but please refer to Spring AI documentation.
# MCP Server Configuration
spring.ai.mcp.server.name=fluxnova-mcp-server
spring.ai.mcp.server.version=1.0.0
spring.ai.mcp.server.protocol=streamable
spring.ai.mcp.server.stdio=false
spring.ai.mcp.server.type=sync
The MCP components should be directly testable via npx @modelcontextprotocol/inspector. After connecting to your Fluxnova server, you can invoke the process via the Tools menu.

Fully working example is now available here, for testing purposes only. finos/fluxnova-examples#21

pieter-schutte · 2026-03-03T14:48:34Z

@pieter-schutte Sharing my thoughts on the plugin approach that we discussed in last week's maintainers call First of all, I really appreciate and excited for all the great AI contributions Natwest is brining to Fluxnova, this is a great win for Fluxnova community to leverage the opensource collaboration

Advantages of enabling capabilities like AI integrations as plugin

plugins hook into engine bootstrap and configurations allowing organizations to enable/disable on demand basis vs them being part of core engine - This is important because organizations might have different stance and custom integration needed especially for things like AI, MCP, Agentic, Authorization, Events etc

Since plugins are configuration based they can be easily configured with maven dependency and spring config annotation that improves the portability across environments

This allows us cleaner separation of Core engine capabilities which is more around BPMN and DMN engines or any capabilities needed for orchestration and automation with extended capabilities like MCP integration, custom security integration etc

This would also allow us to build granular functionality independently with no disruption to core functionality.

So I recommend that this

MCP backend java code integration become a plugin

Modeler code is added as part of Modeler plugin - which does not impact runtime anyways

Hi Harish. Can you please provide an example implementation of an existing plugin

tom-stavert · 2026-03-04T15:43:52Z

mcp-fluxnova-extension/README.md

+    <!-- Spring AI MCP Server -->
+    <dependency>
+        <groupId>org.springframework.ai</groupId>
+        <artifactId>spring-ai-starter-mcp-server-webmvc</artifactId>


Is it necessary to provide this dependency (since it is declared in the extension's pom.xml)? I was able to get it running without it (with just h2 and fluxnova-bpm-spring-boot-starter-webapp) .

tom-stavert · 2026-03-04T16:33:16Z

I was able to add the extension to an existing Fluxnova spring boot application (independent of the example you have provided, Pieter) without any issues. I think this works well as a plugin/extension so would also be interested if we could get an example implementation of a plugin.

tom-stavert · 2026-03-05T10:04:25Z

Having thought a little more about security for MCP functionality (an aspect we are also considering for our upcoming contribution at Scott Logic), I wonder whether it would make sense to abstract the MCP server configuration into a separate parent plugin that would handle enabling MCP server functionality for Fluxnova and any associated authentication/security features. Then, extensions which add tools (like this one) could be added on top of this. For security itself, we have discussed following a similar approach to the authentication for the REST API but curious to hear your thoughts on this.

pieter-schutte · 2026-03-10T16:11:22Z

This PR will soon be closed. A preview of the plugin based implementation available here: https://github.com/finos/fluxnova-bpm-platform/tree/FLUXNOVA-AI-ISSUE-10

Will raise a PR as soon as the new Plugins repo is online.

@finos/fluxnova-maintainers @JamieWhitMac

ISSUE-10 Initial implementation for MCP Start Events. Includes a plug…

0d2e1c5

…in component that will render a new pallete option called MCP Start Event and a corresponding java components that will register MCP components for each MCP Start event, on Startup and process deployment.

pieter-schutte requested review from a team, EdsGianesi, aurelian-valeriu-apostol-db and mr-dibbs February 11, 2026 10:14

EdsGianesi reviewed Feb 11, 2026

View reviewed changes

pieter-schutte commented Feb 12, 2026

View reviewed changes

HarishMalavade reviewed Feb 13, 2026

View reviewed changes

HarishMalavade reviewed Feb 17, 2026

View reviewed changes

bryankenneally reviewed Feb 19, 2026

View reviewed changes

This was referenced Feb 24, 2026

24 FEB 2026 - Fluxnova AI Working Group Meeting Minutes #16

Open

03 MAR 2026 - Fluxnova AI Working Group Meeting Minutes #18

Open

tom-stavert reviewed Mar 4, 2026

View reviewed changes

		@@ -0,0 +1,58 @@
		'use strict';

		var registerBpmnJSPlugin = require('camunda-modeler-plugin-helpers').registerBpmnJSPlugin;

Conversation

pieter-schutte commented Feb 11, 2026

Uh oh!

EdsGianesi left a comment

Choose a reason for hiding this comment

Uh oh!

pieter-schutte commented Feb 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pieter-schutte Feb 12, 2026

Choose a reason for hiding this comment

Uh oh!

HarishMalavade Feb 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

kriswest Feb 13, 2026

Choose a reason for hiding this comment

Uh oh!

HarishMalavade Feb 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

HarishMalavade left a comment

Choose a reason for hiding this comment

Uh oh!

PieterSchutte commented Feb 13, 2026 • edited by pieter-schutte Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

kziemski commented Feb 17, 2026

Uh oh!

prajwolbhandari1 commented Feb 17, 2026

Uh oh!

HarishMalavade Feb 17, 2026

Choose a reason for hiding this comment

Uh oh!

pieter-schutte commented Feb 18, 2026

Uh oh!

HarishMalavade commented Feb 18, 2026

Uh oh!

bryankenneally Feb 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

bryankenneally Feb 19, 2026

Choose a reason for hiding this comment

Uh oh!

bryankenneally Feb 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

bryankenneally Feb 19, 2026

Choose a reason for hiding this comment

Uh oh!

prajwolbhandari1 commented Feb 24, 2026

Uh oh!

pieter-schutte commented Feb 25, 2026

Uh oh!

pieter-schutte commented Feb 25, 2026

Uh oh!

bryankenneally commented Feb 25, 2026

Uh oh!

pieter-schutte commented Feb 26, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Spring MCP Properties

Uh oh!

HarishMalavade commented Mar 3, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pieter-schutte commented Mar 3, 2026

Spring MCP Properties

Uh oh!

pieter-schutte commented Mar 3, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

tom-stavert Mar 4, 2026

Choose a reason for hiding this comment

Uh oh!

tom-stavert commented Mar 4, 2026

Uh oh!

pieter-schutte commented Feb 12, 2026 •

edited

Loading

HarishMalavade Feb 13, 2026 •

edited

Loading

HarishMalavade Feb 13, 2026 •

edited

Loading

PieterSchutte commented Feb 13, 2026 •

edited by pieter-schutte

Loading

bryankenneally Feb 19, 2026 •

edited

Loading

bryankenneally Feb 19, 2026 •

edited

Loading

pieter-schutte commented Feb 26, 2026 •

edited

Loading

HarishMalavade commented Mar 3, 2026 •

edited

Loading

pieter-schutte commented Mar 3, 2026 •

edited

Loading

tom-stavert commented Mar 5, 2026 •

edited

Loading