How to move agentic AI from pilot to production, without fighting your own infrastructure
MuleSoft announces a host of new capabilities for building and running agentic workflows, giving IT and platform teams the tools to discover every AI asset across their environment, make existing APIs and agents agent-ready, orchestrate them reliably at scale, and manage it all from wherever their teams already work. Agent Scanners, MCP Bridge, and Agent Broker are generally available now; A2A Bridge, MuleSoft Playground, and Agentic Mule App Upgrades arrive later this quarter.
–
97% of IT leaders report facing challenges with their agentic transformation initiatives.³ Counterintuitively, agents were meant to simplify this work. Without the right foundation, they often do the opposite: 86% of IT leaders warn that without proper integration, AI agents add more complexity than value.³
This isn’t a shortcoming of the technology — AI and agents are capable of real value. The issue is the environment they’re dropped into, and it consistently comes down to five problems:
- No central inventory of what exists across the environment
- APIs and assets that aren’t built for agents to discover or trust
- Agents that can’t work together at scale
- Even connected agents that don’t know what data to trust
- Tools scattered across consoles teams don’t actually live in
Five points of friction, all leading back to the same thing: complexity, compounding at every layer. Here’s how MuleSoft’s latest innovations take on each one — starting with visibility.
Find Any Agent, Regardless of Where It’s Deployed
Agent Scanners
The first barrier is visibility: you can’t leverage what you can’t see. Agent Scanners, already generally available, are how MuleSoft solves that.
Scanners are the auto-population mechanism behind the Agent Fabric Registry, MuleSoft’s catalog for agents and AI assets. Instead of requiring developers to manually register every agent they build — a process that creates a governance gap the moment someone forgets a step — Scanners continuously detect new and updated agents, MCP servers, and APIs on a schedule, and bring them into the registry automatically.
What’s new: expanded scanning support across more of the ecosystems organizations are already using, including Databricks, Snowflake, Kong, Apigee, Agentforce, Claude, and Langsmith, alongside the providers already supported like Amazon, Azure, Google, and more. Wherever agents, MCP servers, and APIs are running, Scanners can find them — turning agent sprawl into a single, continuously updated source of truth instead of a static, manually maintained list.
Make Your Existing Environment Agent-Ready
MCP Bridge
The second barrier: most APIs and AI assets aren’t agent-ready. They exist, they work fine for humans, but today, agents can’t discover or call most APIs without custom code. MCP Bridge, already generally available, closes that gap without requiring anyone to rebuild what they already have.
MCP Bridge transforms existing REST API investments into MCP servers in just a few clicks, so agents can discover and use them through a standardized protocol — no rewriting services, no new runtimes, no weakening of the governance already in place. It works as an interfacing layer at the same API gateway layer enterprises already use to manage traffic, security, and policy, so existing authentication, rate limiting, and observability continue to apply to agent traffic just as they do to any other consumer.
What’s new: SaaS Actions. Instead of building every MCP tool from scratch, teams get prebuilt, trusted actions for common SaaS applications, natively available inside MCP Bridge — addressing gaps in enterprise-built APIs with accelerated, ready-made SaaS actions. There are currently more than 1000 prebuilt SaaS Actions for over 100 systems available, covering tools like Asana, Jira, HubSpot, Mailchimp, Zendesk, and more.
A2A Bridge
MCP Bridge solves agent-to-system communication. But agents don’t just need to reach systems — increasingly, they need to reach each other. That’s where A2A Bridge (GA Q3 ‘26) comes in.
A2A is an agent interoperability protocol: it details how one agent should work with another. When two agents both speak A2A, they can share information and hand off work seamlessly, because they already share instructions for how to communicate.
The problem is, not every agent speaks it. Agentforce, for example, doesn’t support A2A — so if an organization has a Copilot agent that needs to talk to an Agentforce agent, that doesn’t happen automatically.
With A2A Bridge, teams put the bridge in front of any agent, and it becomes A2A-enabled in a matter of clicks — the same underlying principle as MCP Bridge, applied to agent-to-agent communication instead of agent-to-system.
Make Every Agent Work as One System
Agent Broker
The third barrier: agents don’t work together seamlessly. Even once agents exist, getting them to communicate and act together — securely, at scale — is its own unsolved problem. Agent Broker, already generally available, is how MuleSoft solves it.
Agent Broker is a context-aware routing service that discovers and engages the best-fit agents and tools for the job, so complex, cross-platform processes run seamlessly and predictably — no matter where each agent was built.
What’s new: Guided Determinism with Agent Script, Salesforce’s open scripting language that blends deterministic, fixed-logic workflows with agentic reasoning. Where a workflow needs a guaranteed, repeatable outcome, teams get fixed logic. Where they want the agent to adapt, they let it reason. Same broker, same agent network — just dialing in how much control versus flexibility each step needs.
Here’s what that looks like in practice: imagine an IT support team building a broker to investigate and resolve support tickets. Some tickets are high severity — say, unauthorized login attempts on a user’s account — and need to be routed to the on-call team immediately, every time, with no ambiguity. Most tickets, though, are low severity, better suited to LLM reasoning and specialized agents, like a Help Center Agent or a License Provisioning Agent, working together to investigate root cause and resolve the issue. Guided determinism means defining the business-critical parts of that flow with deterministic logic, while giving AI full reasoning power everywhere it’s safe to do so: guaranteed control where it matters, and AI-powered flexibility where it isn’t.
Work From Wherever Your Team Already Lives
Headless MuleSoft
The fourth barrier: switching tools slows teams down. The consoles used to manage all of this typically live in one place. Teams don’t — they’re in Slack, in Microsoft Teams, in the apps they already use every day.
It starts with a simple question: why should a team have to leave the tools they already use, just to get work done?
The platform now has a refreshed interface and an embedded AI agent teams can simply talk to. But underneath both is headless MuleSoft — separating the platform’s power from its screen, because no user should be tied to a specific UI to get work done.
At the center of this is the MuleSoft Platform MCP Server, a remote MCP server that lets any MCP-compatible client — Claude Desktop, Windsurf, Cursor, VS Code, or a team’s own agent — connect directly to MuleSoft. Once connected, developers and agents can work with their APIs, agents, MCP servers, LLMs, governance data, and monitoring information without switching to the Anypoint UI. A developer can simply ask their AI tool, “Which APIs do we already have for customer onboarding?” or “Show me the governance status for this API,” without leaving the tool they’re already using. MuleSoft’s headless capabilities turn the tools a team already uses into a direct control layer for the platform — and critically, the permission model and governance stay the same no matter which surface someone works from.
So teams get the best of both: a modern interface, or managing everything directly from Slack, Microsoft Teams, Claude, Cursor, or their own tools. Every capability, delivered wherever teams already live.
One Place to Test: MuleSoft Playground
Access isn’t just about managing the platform — it’s also about understanding what’s built on it. Today, teams often jump between three or four separate tools just to see how a service works. By the time everything’s configured, it’s easy to lose track of what you were trying to access. That friction stalls deployments.
MuleSoft Playground (GA Q3 ‘26) changes that. It gives teams a single discovery experience for APIs, MCP servers, LLM proxies, and agents, built directly into the platform. Regardless of where a service lives or is deployed, teams can connect to it, explore it’s capabilities, and see how it behaves – all from within MuleSoft.
With one place to try API andAI assets, teams can eliminate unnecessary app switching, validate functionality in minutes instead of hours, and quickly understand what each asset does – accelerating development and deployment.
The API and MCP Server Playgrounds launch in the coming weeks, followed by LLM Proxy Playground and Agent Playground come next.
Maintenance Doesn’t Stop at Launch: Agentic Mule App Upgrades
Building agentic workflows doesn’t stop at launch. Every Mule app depends on two things underneath it: Java and the Mule Runtime, and historically, keeping those current has been a slow, manual process — updating versions, checking connector compatibility, fixing what breaks, re-testing, one app at a time.
Agentic Mule App Upgrades (GA Q3 2026) automates that entire process with MuleSoft Vibes. It’s an AI agentic experience that handles the upgrade end to end — accelerating Java and runtime upgrades and lowering the ongoing cost of maintaining Mule applications.
Give Agents Data They Can Actually Trust
MuleSoft + Informatica
Everything above solves for discovery, readiness, and orchestration of agents and APIs. But agents are only as trustworthy as the data underneath them — which is where MuleSoft’s co-innovation with Informatica comes in, giving organizations a platform-agnostic foundation for trusted enterprise data.
Agent Fabric Context Catalog (GA Q3 2026) brings Informatica’s data catalogs directly inside MuleSoft, so teams can see the full hierarchy of how data connects to the agents and MCP servers in their workflows — turning governance into something teams can see and act on, rather than reconstruct after the fact.
The Informatica MDM Connector (GA now) gives teams a ready-built, point-and-click way to connect Mule applications to Informatica MDM, no custom code or API expertise required, so trusted, high-quality master data flows across agents and systems faster, with less effort and fewer errors.
Informatica MCP Servers give agents secure, ready-made access to trusted, governed data and data operations, without anyone having to build that connection themselves.
Two are GA today: the CDGC Metadata Search MCP Server lets agents search metadata and run compliance checks, powered by CLAIRE, Informatica’s own agent. The Cloud Address Verification MCP Server lets agents validate, standardize, and enrich global addresses in real time.
Three more arrive in August: a Data Provisioning MCP Server, so agents can discover and request access to governed data collections; a Customer Identification MCP Server, so agents can resolve identities and pull trusted master records; and a Job Management MCP Server, so agents can orchestrate and monitor data integration jobs in real time.
Together, this means agents can securely discover, access, and act on enterprise data — not just enterprise APIs.
The Takeaway
The organizations that act now are the ones that will be shipping agentic AI while their competitors are still stuck in pilot. MuleSoft gives teams a direct way to do that: discover every agent and asset across the environment, make existing APIs and agents ready to work with AI, orchestrate everything reliably at scale, and give teams access from wherever they already work.
- Discoverability closes the visibility gap — Agent Scanners bring every agent, MCP server, and API into a single, continuously updated registry.
- Agent-readiness closes the readiness gap — MCP Bridge and A2A Bridge make existing APIs and agents discoverable and interoperable without rebuilding them, and Informatica’s co-innovations extend that same trust to the data underneath.
- Orchestration closes the coordination gap — Agent Broker routes work to the best-fit agents reliably, with guided determinism where it matters most.
- Access closes the “your tools live in one console, your teams don’t” gap — headless MuleSoft, MuleSoft Playground, and Agentic Mule App Upgrades all mean teams can build, test, and maintain agentic workflows from wherever they already work.
That’s the shift from “everyone’s experimenting” to organizations actually shipping — not by asking teams to do more, but by easing the unnecessary complexity that was standing in their way.
Sources
- Deloitte, State of AI in the Enterprise (2026)
- Larridin, State of Enterprise AI (2025)
- 2026 MuleSoft Connectivity Benchmark Report, with insights from Deloitte Digital
