March 2026 Product Club | MCP Inbound Support

Welcome to the March Product Club Recap!

This month, the Automation Anywhere community got an in-depth look at MCP Inbound and what it means for teams who want their AI agents to securely trigger enterprise automations without building custom integrations from scratch. The session featured Rajendra Vijay (Director of Product Management, Agent Interoperability) and Max Cassidy (Senior Developer Evangelist), who walked through the core concepts, a live demo, and a forward-looking roadmap for multi-agent orchestration.

From APIs to a Standard: What MCP Actually Solves

Rajendra opened by grounding the session in a real developer pain point: as AI agents have become central to enterprise workflows, connecting them to business systems has meant custom API integration work every single time, each with its own authentication scheme, data format, and maintenance burden. MCP (Model Context Protocol) changes that equation. Instead of building point-to-point connections, teams install one MCP client and interact with any MCP-compliant server through a common layer. The result: agents can dynamically discover what automations are available, trigger them in plain language, and do so within enterprise-grade security and governance guardrails.

Seeing It in Action

Max picked up the story from last month where webhooks were used to trigger a new hire workflow and added a new layer: deploying that same automation on demand through a chat interface, no technical background required. The demo showed an IT support agent opening a queue of 50+ tickets and simply typing "process my open tickets" into a chat widget. Behind the scenes, five steps happened automatically: the request hit the MCP server, credentials were exchanged for a secure session token, the MCP server instructed Control Room to deploy the automation, and the bot began processing every ticket in the queue: classifying, routing, and actioning each one.

Max's analogy made the architecture click: think of MCP like air traffic control. The pilot (the user) states their intent. The tower (MCP) translates that into coordinated instructions. The ground crews and runway systems (Automation Anywhere) execute the work. One statement with many systems and no manual handoffs.

The broader implication is significant: this same pattern applies to any business process. An HR manager saying "onboard these three new hires starting Monday." A procurement lead saying "process all pending purchase orders under $10,000." A compliance officer saying "run the quarterly access review." Each of these can trigger end-to-end automated workflows without anyone needing to know which bot to run or how to configure it.

The Vision: End-to-End Multi-Agent Orchestration

Rajendra closed with a vision demo showing where the platform is headed. The scenario followed Marcus, an AI factory lead, managing a quote generation process with some steps already automated, and others requiring the kind of variability and exception handling that only agentic AI can navigate. The demo showed outbound agent connections being configured so that Automation Anywhere's goal-based AI agents could call a Salesforce AI agent for catalog data, with administrators controlling exactly which skills are exposed, which users have access, and which capabilities are authorized. That last point matters: Rajendra called out tool poisoning attacks as a real enterprise concern, and the granular approval model is a direct response to it.

From there, a co-pilot for automators dropped the updated AI agent into an existing Mozart Orchestrator process automatically. No drag and drop. No manual searching. The completed process was then made available to Microsoft Copilot Studio via the MCP server, and a business user in Microsoft Teams uploaded a document and asked for a quote in plain language. The full multi-agent workflow ran in the background, and the response came back without the user ever leaving Teams.

What's Available Now and What's Coming

A few important markers for teams evaluating or actively building:

• Available today (A3.9): MCP inbound support for cloud customers — securely discover and trigger task bots, API Tasks, and processes from any MCP-compliant AI assistant. Attended automation is supported, with role-based access control and full audit logging of every tool call.
• Next release (A4.0): OAuth 2.1 support (replacing static API keys), unattended automation on remote devices, the ability to trigger Automation Anywhere goal-based AI agents, improved variable descriptions, and better session state management.
• On the roadmap: MCP outbound (goal-based agents calling third party systems), on-prem support (beginning with private cloud), A2A (agent-to-agent) orchestration, Community Edition support, and the ability to invoke custom scripts and API calls from within Automation Anywhere.

Licensing Note

MCP inbound is available to all cloud customers. Advanced capabilities, including Process Reasoning Engine integration and AI governance, require an Enterprise (or AP Pro/AP Essential) license. On-prem support is planned after the A4.1 release.

Swag Winners

As promised, we randomly selected some swag winners from our engaged audience. Join me in celebrating:

​@Havva 

​@rewilliams 

​@Ajaysh81 

Session Q&A

Q: What licenses are required to use MCP inbound? A: MCP inbound is available to all cloud customers. Advanced capabilities — such as Process Reasoning Engine integration, AI governance, and pre-built static tools — require an Enterprise, AP Pro, or AP Essential license. On-prem support is planned after the A4.1 release.

Q: Are inbound connections available today? What about outbound? A: Yes — inbound connections are available today in A3.9 for all cloud customers. Outbound connections, which allow Automation Anywhere goal-based AI agents to call third-party systems, are currently on the roadmap and in active development.

Q: How does MCP work for customers who can't use the Automation Anywhere Credential Vault and need to use their own enterprise IDP? A: The way your automations run today doesn't change. If a human manually enters credentials mid-automation (attended), that continues to work with MCP inbound. If bots fetch credentials from a third-party vault via API calls, that also continues to work. MCP inbound doesn't alter how credentials are sourced or managed within your automations.

Q: Can MCP be used with a base license, or is an additional AI license required? A: Cloud customers can access MCP inbound capabilities on the base license. Some advanced capabilities require an Enterprise or higher license. The team noted that licensing details are still being refined, so reaching out directly for specifics based on your current setup is recommended.

Q: How does the setup work with Microsoft 365 Copilot specifically? A: Step-by-step configuration instructions for connecting Microsoft Copilot to the Automation Anywhere MCP server are available in the product documentation. For environment-specific questions or a tailored walkthrough, the team encouraged attendees to request a follow-up session directly.

Q: Can on-prem customers use MCP inbound? A: Not yet. On-prem support is planned after the A4.1 release, starting with private cloud deployments. The team is actively scoping the solution.

For more information on this topic, check out ​@Rajendra Vijay’s blog post on why old integration patterns fail here: LINK

Next month's Product Club on April 8th will build on this same environment, highlighting split and merge capabilities within Mozart Orchestrator — the final chapter of this quarter's workflow. We'll see you there.