Agent-to-agent connections, built-in system tools, Arabic, and a redesigned Studio

This is a major release. MagOneAI now speaks the Agent2Agent (A2A) protocol, so your agents and workflows can work with AI agents running on other platforms, in both directions. Eight built-in system tools cover everyday workflow steps, like waiting, logging or validating data, without an agent or any code. Studio is now available in Arabic with a full right-to-left layout, and Studio and Hub have been redesigned, with new navigation, dark mode and a much more capable canvas. You can also compare versions of a workflow, sign in with OAuth to many popular tool servers, and track spend and reliability on rebuilt analytics dashboards. Alongside these, this release includes a large set of security hardening changes and bug fixes.

Highlights

Agent2Agent (A2A) in both directions

An agent is an AI assistant that can reason, use tools and complete tasks. Agent2Agent (A2A) is an open protocol that lets agents built on different platforms discover each other and hand work to one another, much like an API does for ordinary software.

MagOneAI now supports A2A both ways:

• Share your agents. Any workflow or library agent can be exposed to outside systems as an A2A skill. Each one is described by a signed agent card, a machine-readable profile that tells other agents what it does and proves it really comes from you. Partners call it using scoped partner tokens, access keys that only unlock what you've chosen to share.
• Call outside agents. A new A2A Agent node on the canvas lets a workflow call a registered remote A2A agent as one of its steps. ORBY, the Studio assistant that builds workflows from a chat, can add and wire up this node on its own.

Why it matters: Your MagOneAI agents no longer have to work in isolation. You can offer them to partners and other teams in a standard way, with control over who can use what, and you can bring specialist agents from other platforms into your own workflows.

Learn more: Workflows overview · Build with ORBY

Eight built-in system tools for Tool nodes

In MagOneAI, a workflow (also called a use case) is a series of steps, called nodes, laid out on a visual canvas. A Tool node runs one specific action directly, without an AI model deciding what to do.

Tool nodes now come with eight built-in system tools that need no agent, no code and no external tool server:

• Stop and Error ends the run with an error you define.
• Wait pauses the workflow for a set time.
• Log writes a message to the run's log.
• Set Variables stores values for later steps to use.
• Redact removes sensitive values from data.
• Validate checks that data has the shape and values you expect.
• Transform reshapes data from one format to another.
• Notify sends a notification.

Why it matters: Common housekeeping steps that used to need an agent, a Code node or a separate tool server can now be dropped straight onto the canvas. Workflows get simpler, faster and more predictable.

Learn more: Tool node · Tools overview

Arabic language support

Studio is now available in Arabic, with a full right-to-left (RTL) layout so menus, panels and text read naturally. A new language switcher lets each person choose the language they work in.

Why it matters: Arabic-speaking teams can build and manage workflows in their own language, without fighting a layout designed for left-to-right text.

Learn more: MagOneAI Studio

A redesigned Studio and Hub, with dark mode

Studio is where you build agents and workflows; Hub is where people use them. Both have been redesigned:

• New navigation with a compact icon rail and a project sidebar, so you can move between areas and projects quickly.
• Dark mode, plus a choice of colour palettes in settings.
• Project icons and colours, so projects are easy to tell apart at a glance.
• The Agent Library now opens as full pages instead of side panels.

Why it matters: The everyday screens are cleaner and quicker to get around, and you can set them up to suit how and where you work.

Learn more: MagOneAI Studio · MagOneAI Hub

Canvas upgrades

The canvas is the visual editor where you lay out a workflow's nodes and connect them. It gains a set of tools for building and tidying larger workflows:

• Notes pinned to nodes, so explanations stay next to the step they describe.
• Automatic layout, which arranges your nodes neatly in one click.
• Nodes coloured by role, so it's easy to see at a glance which steps are agents, tools, branches and so on.
• Alignment tools for lining up nodes by hand.
• A command palette for finding actions by typing.
• Pointer and hand modes, for selecting nodes or panning around the canvas.
• A horizontal or vertical layout toggle, so workflows can flow left-to-right or top-to-bottom.

Why it matters: Large workflows are easier to read, explain and maintain, and less time goes into dragging boxes around.

Learn more: Workflows overview

Compare workflow versions

Every time you save a workflow, MagOneAI keeps a revision you can go back to. You can now compare any two saved revisions, either side by side on the canvas or as a JSON diff showing exactly what changed.

Why it matters: Before you publish or roll back a change, you can see exactly what's different, which makes reviews quicker and mistakes easier to catch.

Learn more: Versions and publishing

Sign in with OAuth to remote tool servers

Tools let agents act in other systems. Many tools reach MagOneAI through MCP servers, tool servers that use the Model Context Protocol, an open standard for connecting AI agents to outside tools. OAuth is the familiar "Sign in with…" flow that lets you grant access to an app without sharing your password.

You can now sign in with OAuth to remote MCP servers for Notion, ClickUp, Calendly, Trello, Contentful, Explorium, Dropbox, Slack, Asana and Box. There are also featured one-click cards for popular servers such as GitHub, Hugging Face, Stripe, Supabase and Firecrawl. The Oracle Fusion MCP server now supports Oracle Fusion OAuth as well.

Why it matters: Connecting agents to the tools your teams already use takes a couple of clicks, with no API keys to create, copy or rotate.

Learn more: External MCP servers · MCP credentials

Rebuilt analytics dashboards

The organization and project analytics dashboards have been rebuilt. They now show spend, reliability, items that need your attention and a heat map of when runs happen. The Hub dashboard is now personal and action-first, showing what matters to you and what you need to do next.

Why it matters: It's easier to see what your AI workflows cost, how reliably they run, and where to step in, without digging through individual runs.

Learn more: Analytics · MagOneAI Hub

Improvements

• The Decision node is now the Intelligent Router. The node that asks a decision model questions about your data has a new name and a cleaner properties panel. If the decision model it uses has been removed or made inactive, the node is flagged and the workflow can't be saved until you pick another one. See Decision node.
• Readable URLs. Projects, knowledge bases and use cases now use readable names (slugs) in their web addresses instead of long IDs, so links are easier to recognise and share.
• Simpler tool connections. Each account now has one connection per MCP integration. Jira and Confluence connections remember the site you picked when you signed in, and Outlook Email connections show the shared mailboxes you can use.
• Tool calling with newer OpenAI models. GPT-6 and newer OpenAI models can now call tools while reasoning is turned on. Model configurations also declare whether a model supports tool calling, and you can test it.
• API poll tokens last while a run waits on a person. When a run is paused on a human task, the token used to check its progress through the API now stays valid, instead of expiring after 24 hours.
• Workspace search for superadmins. Workspace pickers in the superadmin portal now search across all workspaces, not just the first 100.
• Retired node types fail clearly. The old Respond and Router node types have been retired. Workflows that still contain them now show a message naming the node to use instead, rather than failing part-way through a run.

Bug fixes

• Decision-only models can be added from the UI. Test Model now runs the decision check, so models used only for decisions pass the test.
• Tool-calling agents on GPT-6 work again, instead of failing with a reasoning_effort error.
• A2A nodes run correctly in production, no longer failing with an "A2A client is disabled" error, and the Agent to Agent tab now shows for organization owners.
• Superadmin workspace pickers no longer retry a failed page over and over.
• Canvas fixes. A sideways trackpad swipe no longer navigates your browser back or forward, and branch handles are shown in full.
• Redesign polish. Project cards show each project's run health, the top bar lays out properly on smaller screens, and a number of other Studio and Hub issues are fixed.

Security

• Stronger sign-in protection. Accounts are locked after repeated wrong passwords, reused refresh tokens (the long-lived tokens that keep you signed in) are detected, and the platform refuses to start with the default secret key.
• No more open redirects on login. The Hub and Studio login pages no longer follow an untrusted ?next= address, which could have been used to send people to a malicious site after signing in.
• Test results respect permissions. Evaluating a test case no longer reveals the output of runs you aren't allowed to see.
• Platform OAuth secrets stay protected. External MCP servers can no longer declare their own OAuth settings to obtain the platform's client secrets.
• A hardened Code node sandbox. The isolated environment that runs Code node scripts has been tightened, and its temporary storage behaves the same on every deployment.