From 3c0bec565b2ac7bbe1a319bc548ce2a23bbdf53b Mon Sep 17 00:00:00 2001 From: Dimitrie Hoekstra Date: Mon, 24 Aug 2026 11:40:30 +0200 Subject: [PATCH 1/9] changelog: connect your own AI agent, with the 3.0 feature catalog entry --- nuxt/content/feature-catalog.yml | 8 +++ .../2026/08/connect-your-own-ai-agent.md | 26 +++++++++ .../2026/08/images/connect-your-own-agent.svg | 55 +++++++++++++++++++ 3 files changed, 89 insertions(+) create mode 100644 src/changelog/2026/08/connect-your-own-ai-agent.md create mode 100644 src/changelog/2026/08/images/connect-your-own-agent.svg diff --git a/nuxt/content/feature-catalog.yml b/nuxt/content/feature-catalog.yml index 043b9f695c..34ec4d8fa9 100644 --- a/nuxt/content/feature-catalog.yml +++ b/nuxt/content/feature-catalog.yml @@ -59,6 +59,14 @@ sections: beta: true showOnPricing: false tiers: { edge: true, hub: true, fleet: true } + - id: third-party-agents + title: Connect Your Own AI Agent + description: "Point the AI agent your company already uses at FlowFuse, so it can query your teams and instances and build Node-RED applications, with you deciding which teams it reaches and whether it can make changes." + docsLink: /docs/user/expert/third-party-agents/ + changelog: + - url: /changelog/2026/08/connect-your-own-ai-agent/ + release: "3.0" + tiers: { edge: true, hub: true, fleet: true } - id: mcp-servers title: Agentic Operations description: "Expose your Node-RED flows as tools an AI agent can call directly, so agents can query the state of the factory or trigger actions without custom integration work." diff --git a/src/changelog/2026/08/connect-your-own-ai-agent.md b/src/changelog/2026/08/connect-your-own-ai-agent.md new file mode 100644 index 0000000000..27ccd42b5f --- /dev/null +++ b/src/changelog/2026/08/connect-your-own-ai-agent.md @@ -0,0 +1,26 @@ +--- +title: Connect your own AI agent to FlowFuse +description: The AI agent your team already uses can now work your platform and build Node-RED applications, with you deciding what it may reach. +date: 2026-08-27 12:00:00 +release: "3.0" +authors: ["dimitrie-hoekstra"] +tags: +- changelog +issues: +- https://github.com/FlowFuse/flowfuse/issues/8192 +--- + +FlowFuse Expert is no longer the only AI that can work your platform. FlowFuse now acts as an MCP server, so the agent your team already uses can query your teams and instances, and build Node-RED applications for you. + +![Your own AI agent signs in to one FlowFuse address, and FlowFuse reaches your instances and your open editor session](./images/connect-your-own-agent.svg) +*Your agent talks to one address. FlowFuse reaches the instances, so they stay unexposed.* + +Add the FlowFuse address in your agent's connector settings, sign in, and choose which teams the agent may reach and whether it may make changes. Microsoft Copilot, ChatGPT and Claude all connect this way, and so do command-line and editor agents such as Claude Code, Cursor and Visual Studio Code. + +This matters most where company policy only permits an approved AI assistant. Until now that meant no AI on the platform at all, because the only way in was our own. Now the assistant your company already sanctioned can do the work. + +The access you grant is enforced on every call, so a read-only grant is refused whatever the agent tries. Nothing an agent reaches through FlowFuse can delete an instance, an application, a snapshot or a team, and deploying stays yours. + +Flow building happens on the canvas in front of you. Open the instance, turn on the MCP toggle in the page header, and you watch the agent work. + +This is available to all FlowFuse Cloud users and to Enterprise self-hosted installations from v3.0, across FlowFuse Hub, Edge and Fleet. See [connecting your own agent](/docs/user/expert/third-party-agents/) for setup, including what to check on an instance before asking for a flow. diff --git a/src/changelog/2026/08/images/connect-your-own-agent.svg b/src/changelog/2026/08/images/connect-your-own-agent.svg new file mode 100644 index 0000000000..9ec26d3f53 --- /dev/null +++ b/src/changelog/2026/08/images/connect-your-own-agent.svg @@ -0,0 +1,55 @@ + + How your own agent reaches FlowFuse + + + + + + + + + + + + + + Your own AI agent + Copilot, ChatGPT, Claude + + + signs in + + + + WHAT YOU GRANTED + + + + One FlowFuse address + which teams it may reach, + and whether it may change them + + + + + + FlowFuse + your platform + + + + + + + Instances + and remote + + + Your editor + when you open it + + + Your agent talks to one address. FlowFuse reaches the instances, so they stay unexposed. + + From 3035b2f5d29102656395816e8ec4553b2eb65b7e Mon Sep 17 00:00:00 2001 From: Dimitrie Hoekstra Date: Mon, 24 Aug 2026 14:22:20 +0200 Subject: [PATCH 2/9] changelog: drop the exposure wording from the diagram and its caption --- src/changelog/2026/08/connect-your-own-ai-agent.md | 2 +- src/changelog/2026/08/images/connect-your-own-agent.svg | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/src/changelog/2026/08/connect-your-own-ai-agent.md b/src/changelog/2026/08/connect-your-own-ai-agent.md index 27ccd42b5f..c20809a56c 100644 --- a/src/changelog/2026/08/connect-your-own-ai-agent.md +++ b/src/changelog/2026/08/connect-your-own-ai-agent.md @@ -13,7 +13,7 @@ issues: FlowFuse Expert is no longer the only AI that can work your platform. FlowFuse now acts as an MCP server, so the agent your team already uses can query your teams and instances, and build Node-RED applications for you. ![Your own AI agent signs in to one FlowFuse address, and FlowFuse reaches your instances and your open editor session](./images/connect-your-own-agent.svg) -*Your agent talks to one address. FlowFuse reaches the instances, so they stay unexposed.* +*Your agent talks to one address. FlowFuse reaches the instances, so they are never opened up.* Add the FlowFuse address in your agent's connector settings, sign in, and choose which teams the agent may reach and whether it may make changes. Microsoft Copilot, ChatGPT and Claude all connect this way, and so do command-line and editor agents such as Claude Code, Cursor and Visual Studio Code. diff --git a/src/changelog/2026/08/images/connect-your-own-agent.svg b/src/changelog/2026/08/images/connect-your-own-agent.svg index 9ec26d3f53..118c1de9d1 100644 --- a/src/changelog/2026/08/images/connect-your-own-agent.svg +++ b/src/changelog/2026/08/images/connect-your-own-agent.svg @@ -50,6 +50,6 @@ when you open it - Your agent talks to one address. FlowFuse reaches the instances, so they stay unexposed. + Your agent talks to one address. FlowFuse reaches the instances, so they are never opened up. From c422fb4e7760828888938c3bfc243684e2ca56e0 Mon Sep 17 00:00:00 2001 From: Dimitrie Hoekstra Date: Mon, 24 Aug 2026 17:41:20 +0200 Subject: [PATCH 3/9] changelog: set the author --- src/changelog/2026/08/connect-your-own-ai-agent.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/src/changelog/2026/08/connect-your-own-ai-agent.md b/src/changelog/2026/08/connect-your-own-ai-agent.md index c20809a56c..73a065319d 100644 --- a/src/changelog/2026/08/connect-your-own-ai-agent.md +++ b/src/changelog/2026/08/connect-your-own-ai-agent.md @@ -3,7 +3,7 @@ title: Connect your own AI agent to FlowFuse description: The AI agent your team already uses can now work your platform and build Node-RED applications, with you deciding what it may reach. date: 2026-08-27 12:00:00 release: "3.0" -authors: ["dimitrie-hoekstra"] +authors: ["serban-costin"] tags: - changelog issues: From cee42caad0c7fe3871df6cffa0c3b31660783bb3 Mon Sep 17 00:00:00 2001 From: Dimitrie Hoekstra Date: Wed, 26 Aug 2026 11:31:06 +0200 Subject: [PATCH 4/9] changelog: give it a CTA that actually installs the agent Per review, the entry ended on a docs link, which is not a call to action. It now carries the one-click routes that exist: the VS Code redirect and the Cursor deeplink, both verified (the VS Code one 302s to vscode:mcp/install with the right payload), the claude mcp add command, and the paste-the-address route for Claude, ChatGPT and Copilot Studio, which have no deeplink. Also: agent rather than assistant, matching the docs, and a line on the audit log now that third-party calls are attributed via MCP. --- src/changelog/2026/08/connect-your-own-ai-agent.md | 13 ++++++++++--- 1 file changed, 10 insertions(+), 3 deletions(-) diff --git a/src/changelog/2026/08/connect-your-own-ai-agent.md b/src/changelog/2026/08/connect-your-own-ai-agent.md index 73a065319d..7a7567a69f 100644 --- a/src/changelog/2026/08/connect-your-own-ai-agent.md +++ b/src/changelog/2026/08/connect-your-own-ai-agent.md @@ -15,11 +15,18 @@ FlowFuse Expert is no longer the only AI that can work your platform. FlowFuse n ![Your own AI agent signs in to one FlowFuse address, and FlowFuse reaches your instances and your open editor session](./images/connect-your-own-agent.svg) *Your agent talks to one address. FlowFuse reaches the instances, so they are never opened up.* -Add the FlowFuse address in your agent's connector settings, sign in, and choose which teams the agent may reach and whether it may make changes. Microsoft Copilot, ChatGPT and Claude all connect this way, and so do command-line and editor agents such as Claude Code, Cursor and Visual Studio Code. +Add the FlowFuse address in your agent's connector settings, sign in, and choose which teams the agent may reach and whether it may make changes. -This matters most where company policy only permits an approved AI assistant. Until now that meant no AI on the platform at all, because the only way in was our own. Now the assistant your company already sanctioned can do the work. +Some agents take it in one click: -The access you grant is enforced on every call, so a read-only grant is refused whatever the agent tries. Nothing an agent reaches through FlowFuse can delete an instance, an application, a snapshot or a team, and deploying stays yours. +- Visual Studio Code: [add FlowFuse](https://insiders.vscode.dev/redirect/mcp/install?name=flowfuse&config=%7B%22type%22%3A%22http%22%2C%22url%22%3A%22https%3A%2F%2Fapp.flowfuse.com%2Fmcp%22%7D) +- Cursor: [add FlowFuse](cursor://anysphere.cursor-deeplink/mcp/install?name=flowfuse&config=eyJ0eXBlIjoiaHR0cCIsInVybCI6Imh0dHBzOi8vYXBwLmZsb3dmdXNlLmNvbS9tY3AifQ==) +- Claude Code: `claude mcp add --transport http flowfuse https://app.flowfuse.com/mcp` +- Claude, ChatGPT and Copilot Studio: add `https://app.flowfuse.com/mcp` as a custom connector + +This matters most where company policy only permits an approved AI agent. Until now that meant no AI on the platform at all, because the only way in was our own. Now the agent your company already sanctioned can do the work. + +The access you grant is enforced on every call, so a read-only grant is refused whatever the agent tries. Nothing an agent reaches through FlowFuse can delete an instance, an application, a snapshot or a team, and deploying stays yours. Everything it does lands in the audit log, marked `via MCP`. Flow building happens on the canvas in front of you. Open the instance, turn on the MCP toggle in the page header, and you watch the agent work. From 4e5ab2dcdfbbda8ec2d774958db91c9a2e6fe934 Mon Sep 17 00:00:00 2001 From: Dimitrie Hoekstra Date: Wed, 26 Aug 2026 11:41:58 +0200 Subject: [PATCH 5/9] changelog: lead the CTA with the approved-agent route, not developer tooling The previous commit put editor and command-line deeplinks at the top, which inverts the story. The audience is the agent a company has already approved, and /ai leads on the enterprise assistants and local models with coding agents as a trailing note. The CTA now matches that order and hands off to /ai, which already carries the copyable address and the three steps per agent. --- src/changelog/2026/08/connect-your-own-ai-agent.md | 7 +------ 1 file changed, 1 insertion(+), 6 deletions(-) diff --git a/src/changelog/2026/08/connect-your-own-ai-agent.md b/src/changelog/2026/08/connect-your-own-ai-agent.md index 7a7567a69f..992dc0d535 100644 --- a/src/changelog/2026/08/connect-your-own-ai-agent.md +++ b/src/changelog/2026/08/connect-your-own-ai-agent.md @@ -17,12 +17,7 @@ FlowFuse Expert is no longer the only AI that can work your platform. FlowFuse n Add the FlowFuse address in your agent's connector settings, sign in, and choose which teams the agent may reach and whether it may make changes. -Some agents take it in one click: - -- Visual Studio Code: [add FlowFuse](https://insiders.vscode.dev/redirect/mcp/install?name=flowfuse&config=%7B%22type%22%3A%22http%22%2C%22url%22%3A%22https%3A%2F%2Fapp.flowfuse.com%2Fmcp%22%7D) -- Cursor: [add FlowFuse](cursor://anysphere.cursor-deeplink/mcp/install?name=flowfuse&config=eyJ0eXBlIjoiaHR0cCIsInVybCI6Imh0dHBzOi8vYXBwLmZsb3dmdXNlLmNvbS9tY3AifQ==) -- Claude Code: `claude mcp add --transport http flowfuse https://app.flowfuse.com/mcp` -- Claude, ChatGPT and Copilot Studio: add `https://app.flowfuse.com/mcp` as a custom connector +The address is `https://app.flowfuse.com/mcp`. [Pick your agent on the AI page](/ai/) for where that goes in [Microsoft Copilot](https://copilotstudio.microsoft.com/), [ChatGPT](https://chatgpt.com/), [Claude](https://claude.ai/settings/connectors) or [Gemini](https://gemini.google.com/), and for the three steps in each. A local model works the same way through any MCP client that speaks HTTP, and so do command-line and editor agents. This matters most where company policy only permits an approved AI agent. Until now that meant no AI on the platform at all, because the only way in was our own. Now the agent your company already sanctioned can do the work. From 7e6d42dd7159ddd3f3eae6575910599cdb8c859e Mon Sep 17 00:00:00 2001 From: Dimitrie Hoekstra Date: Wed, 26 Aug 2026 11:45:52 +0200 Subject: [PATCH 6/9] changelog: drop the diagram Text only. The schematic stays on /ai, where it sits alongside the picker it explains. --- .../2026/08/connect-your-own-ai-agent.md | 3 - .../2026/08/images/connect-your-own-agent.svg | 55 ------------------- 2 files changed, 58 deletions(-) delete mode 100644 src/changelog/2026/08/images/connect-your-own-agent.svg diff --git a/src/changelog/2026/08/connect-your-own-ai-agent.md b/src/changelog/2026/08/connect-your-own-ai-agent.md index 992dc0d535..3e7d56faae 100644 --- a/src/changelog/2026/08/connect-your-own-ai-agent.md +++ b/src/changelog/2026/08/connect-your-own-ai-agent.md @@ -12,9 +12,6 @@ issues: FlowFuse Expert is no longer the only AI that can work your platform. FlowFuse now acts as an MCP server, so the agent your team already uses can query your teams and instances, and build Node-RED applications for you. -![Your own AI agent signs in to one FlowFuse address, and FlowFuse reaches your instances and your open editor session](./images/connect-your-own-agent.svg) -*Your agent talks to one address. FlowFuse reaches the instances, so they are never opened up.* - Add the FlowFuse address in your agent's connector settings, sign in, and choose which teams the agent may reach and whether it may make changes. The address is `https://app.flowfuse.com/mcp`. [Pick your agent on the AI page](/ai/) for where that goes in [Microsoft Copilot](https://copilotstudio.microsoft.com/), [ChatGPT](https://chatgpt.com/), [Claude](https://claude.ai/settings/connectors) or [Gemini](https://gemini.google.com/), and for the three steps in each. A local model works the same way through any MCP client that speaks HTTP, and so do command-line and editor agents. diff --git a/src/changelog/2026/08/images/connect-your-own-agent.svg b/src/changelog/2026/08/images/connect-your-own-agent.svg deleted file mode 100644 index 118c1de9d1..0000000000 --- a/src/changelog/2026/08/images/connect-your-own-agent.svg +++ /dev/null @@ -1,55 +0,0 @@ - - How your own agent reaches FlowFuse - - - - - - - - - - - - - - Your own AI agent - Copilot, ChatGPT, Claude - - - signs in - - - - WHAT YOU GRANTED - - - - One FlowFuse address - which teams it may reach, - and whether it may change them - - - - - - FlowFuse - your platform - - - - - - - Instances - and remote - - - Your editor - when you open it - - - Your agent talks to one address. FlowFuse reaches the instances, so they are never opened up. - - From 78f6267f3f70a8be85b07af9e79e66d7fd661fdf Mon Sep 17 00:00:00 2001 From: Andrea Palmieri <76187074+andypalmi@users.noreply.github.com> Date: Thu, 27 Aug 2026 12:14:03 +0200 Subject: [PATCH 7/9] Apply suggestion from @andypalmi --- src/changelog/2026/08/connect-your-own-ai-agent.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/src/changelog/2026/08/connect-your-own-ai-agent.md b/src/changelog/2026/08/connect-your-own-ai-agent.md index 3e7d56faae..391beed095 100644 --- a/src/changelog/2026/08/connect-your-own-ai-agent.md +++ b/src/changelog/2026/08/connect-your-own-ai-agent.md @@ -3,7 +3,7 @@ title: Connect your own AI agent to FlowFuse description: The AI agent your team already uses can now work your platform and build Node-RED applications, with you deciding what it may reach. date: 2026-08-27 12:00:00 release: "3.0" -authors: ["serban-costin"] +authors: ["serban-costin", "andrea-palmieri", "stephen-mclaughlin"] tags: - changelog issues: From 997f44af1e498722bff822785437a68f25eea31c Mon Sep 17 00:00:00 2001 From: Dimitrie Hoekstra Date: Thu, 27 Aug 2026 15:21:36 +0200 Subject: [PATCH 8/9] Apply suggestions from code review Co-authored-by: Andrea Palmieri <76187074+andypalmi@users.noreply.github.com> --- src/changelog/2026/08/connect-your-own-ai-agent.md | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-) diff --git a/src/changelog/2026/08/connect-your-own-ai-agent.md b/src/changelog/2026/08/connect-your-own-ai-agent.md index 391beed095..5ebf31f806 100644 --- a/src/changelog/2026/08/connect-your-own-ai-agent.md +++ b/src/changelog/2026/08/connect-your-own-ai-agent.md @@ -10,16 +10,16 @@ issues: - https://github.com/FlowFuse/flowfuse/issues/8192 --- -FlowFuse Expert is no longer the only AI that can work your platform. FlowFuse now acts as an MCP server, so the agent your team already uses can query your teams and instances, and build Node-RED applications for you. +You can now connect your own AI agent to FlowFuse. Your company-approved AI can manage your platform and build Node-RED applications for you. -Add the FlowFuse address in your agent's connector settings, sign in, and choose which teams the agent may reach and whether it may make changes. +Add the FlowFuse address, sign in with OAuth, and you decide which teams the agent reaches and whether it can make changes. -The address is `https://app.flowfuse.com/mcp`. [Pick your agent on the AI page](/ai/) for where that goes in [Microsoft Copilot](https://copilotstudio.microsoft.com/), [ChatGPT](https://chatgpt.com/), [Claude](https://claude.ai/settings/connectors) or [Gemini](https://gemini.google.com/), and for the three steps in each. A local model works the same way through any MCP client that speaks HTTP, and so do command-line and editor agents. +The address is `https://app.flowfuse.com/mcp`. [Pick your agent on the AI page](/ai/) for the exact steps in [Microsoft Copilot](https://copilotstudio.microsoft.com/), [ChatGPT](https://chatgpt.com/), [Claude](https://claude.ai/settings/connectors) or [Gemini](https://gemini.google.com/). Any MCP client that supports the HTTP transport works too, including command-line and editor agents and local models. Sign-in uses OAuth where the client supports it, with a token fallback where it does not. -This matters most where company policy only permits an approved AI agent. Until now that meant no AI on the platform at all, because the only way in was our own. Now the agent your company already sanctioned can do the work. +Your own agent gets the full, growing set of FlowFuse platform automation and flow-building tools to run. That matters most where company policy only permits an approved AI agent, so FlowFuse Expert wasn't an option. -The access you grant is enforced on every call, so a read-only grant is refused whatever the agent tries. Nothing an agent reaches through FlowFuse can delete an instance, an application, a snapshot or a team, and deploying stays yours. Everything it does lands in the audit log, marked `via MCP`. +Your grant is enforced on every call: a read-only agent stays read-only, and no agent can delete an instance, application, snapshot or team. Every action is recorded in the audit log, marked `via MCP`. -Flow building happens on the canvas in front of you. Open the instance, turn on the MCP toggle in the page header, and you watch the agent work. +When the agent builds a flow, you watch it happen on the canvas: open the instance and turn on the MCP toggle in the page header. Nothing goes live until you deploy. This is available to all FlowFuse Cloud users and to Enterprise self-hosted installations from v3.0, across FlowFuse Hub, Edge and Fleet. See [connecting your own agent](/docs/user/expert/third-party-agents/) for setup, including what to check on an instance before asking for a flow. From 2a5a220ed09c36b7a6a91574cc4484e10702fd65 Mon Sep 17 00:00:00 2001 From: Dimitrie Hoekstra Date: Thu, 27 Aug 2026 15:57:35 +0200 Subject: [PATCH 9/9] changelog: drop OAuth from the opening line, it is explained below --- src/changelog/2026/08/connect-your-own-ai-agent.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/src/changelog/2026/08/connect-your-own-ai-agent.md b/src/changelog/2026/08/connect-your-own-ai-agent.md index 5ebf31f806..cd6ba26ab8 100644 --- a/src/changelog/2026/08/connect-your-own-ai-agent.md +++ b/src/changelog/2026/08/connect-your-own-ai-agent.md @@ -12,7 +12,7 @@ issues: You can now connect your own AI agent to FlowFuse. Your company-approved AI can manage your platform and build Node-RED applications for you. -Add the FlowFuse address, sign in with OAuth, and you decide which teams the agent reaches and whether it can make changes. +Add the FlowFuse address to connect, then decide which teams the agent reaches and whether it can make changes. The address is `https://app.flowfuse.com/mcp`. [Pick your agent on the AI page](/ai/) for the exact steps in [Microsoft Copilot](https://copilotstudio.microsoft.com/), [ChatGPT](https://chatgpt.com/), [Claude](https://claude.ai/settings/connectors) or [Gemini](https://gemini.google.com/). Any MCP client that supports the HTTP transport works too, including command-line and editor agents and local models. Sign-in uses OAuth where the client supports it, with a token fallback where it does not.