The Brilliant New Hire Problem

Imagine bringing on the most capable new hire your organisation has ever seen. They understand complex instructions, reason through ambiguous situations, handle multi-step tasks without hand-holding, and can process information faster than any human team. There is just one problem: every system they need to interact with is locked behind a door that requires a developer to open.

They can talk to you. They can plan. They can reason. But they cannot update the CRM, check inventory in the ERP, or pull compliance data from a third-party database - because none of those systems were built to be talked to. They were built for other systems to call, in precise technical formats, following rigid parameter schemas, via custom integrations that each took weeks to build.

That is the integration gap at the heart of enterprise agentic AI in 2026. The AI is ready. The ambition is there. The missing piece is the infrastructure that lets agents reach beyond the conversation window and into the systems where work gets done. That is the problem that MuleSoft Topics and the Model Context Protocol (MCP) exist to close.

Why Your Existing APIs Cannot Serve AI Agents - Yet

Most enterprises have years of meaningful API investment. Integration platforms connect dozens, sometimes hundreds, of systems: ERPs, CRMs, supply chain platforms, compliance databases, and customer portals. On paper, the data and capabilities AI agents need are already exposed as APIs. In practice, those APIs are almost entirely inaccessible to agents without significant custom development work for every new connection.

The reason is structural. Traditional REST APIs were designed for deterministic, code-driven consumption. A calling system sends an exact request in a specified format and receives a predictable response. There is no mechanism for an AI agent to discover what an API can do, understand the business intent behind its operations, or reason about when and how to invoke it within a multi-step workflow.

Before solutions like Topics and MCP emerged, bridging this gap meant building bespoke integrations for every agent-to-system connection - expensive, slow, and brittle when either the agent or the underlying system changed. Data readiness and integration challenges are cited by 35% of enterprises as the top obstacle to scaling agentic AI - above technology limitations, budget constraints, and even talent gaps. The bottleneck is not the AI. It is the infrastructure connecting it to the systems where work actually happens.

The Discoverability Layer: Packaging APIs as Agent Actions

MuleSoft for Agentforce: Topic Center directly addresses the discoverability problem. Now generally available, it allows developers to package existing APIs into three components that together transform an endpoint into something an AI agent can genuinely understand and use.

Actions are the specific tasks the agent can perform - the API operations surfaced for agent consumption. An order management API might expose actions for checking status, updating a delivery address, or initiating a return. A compliance database might expose actions for validating commodity codes against a regulatory reference list.

Instructions are the natural language guidance that tells the agent how and when to use those actions in context - not just what the API can do, but the reasoning to apply: when to call this action, what conditions should be met beforehand, how to interpret the response.

Topics bundle related Actions and Instructions into a coherent, governable unit that an agent can be assigned in Agent Builder.

The flow is unified on a single platform: design the Topic in Anypoint Code Builder, publish it to the API Catalog, and a Salesforce admin assigns it to an agent in Agent Builder - without writing custom integration code at any stage. Developers define the Topic once; admins deploy it without needing to re-engage the development team for each new agent use case. The division of responsibility matters as much as the technical capability.

MCP: The Open Standard That Makes Any API Agent-Ready

MuleSoft Topics solve the agent-readiness problem within the Salesforce ecosystem. The Model Context Protocol (MCP) solves it across the entire enterprise technology landscape.

MCP is an open standard that defines how AI models connect to external tools, data sources, and systems. The analogy that has stuck in the industry is accurate: MCP is the USB-C of AI integrations. Before USB-C, every device needed its own proprietary cable. Before MCP, connecting an AI agent to ten business tools required ten custom integrations per AI model - an exponential maintenance burden that made enterprise agentic AI impractical at scale. MCP standardises that connection so each tool needs one server that works with every MCP-compatible agent.

The adoption velocity tells its own story. MCP grew from roughly 2 million downloads at launch to 97 million monthly in just 16 months - one of the fastest open-source protocol adoption curves in history. Kubernetes, now considered foundational cloud infrastructure, took nearly four years to reach comparable density. Every major AI provider - Anthropic, OpenAI, Google, Microsoft, and AWS - now supports it. In December 2025, Anthropic donated MCP to the newly formed Agentic AI Foundation under the Linux Foundation, co-founded with OpenAI and Block, with Amazon, Google, Microsoft, Cloudflare, and Bloomberg as platinum members. MCP is no longer one company's protocol. It is industry infrastructure.

With MuleSoft's Anypoint Platform now supporting MCP in general availability, organisations can convert any existing API or Mule application into an MCP-compliant server. That means an entire integration estate built over years becomes consumable by any MCP-compatible agent - a configuration exercise rather than a development project.

Governance is built into the same layer. MuleSoft Flex Gateway provides MCP support with allowlists, rate limiting, identity management, and per-agent access controls - so making APIs agent-readable does not mean making them open to every agent that connects. Admins control exactly which tools and metadata are exposed to which agents, keeping the agentic access model granular, auditable, and reversible.

The Starting Point is Already in Your API Estate

The practical barrier to getting started is lower than most enterprises assume. The API estate an organisation has already built is not a legacy liability to be replaced. It is the asset that becomes agent-ready through Topic definition and MCP server configuration.

The starting point is a focused inventory: which existing APIs connect to the systems where agents will need to act? What is the business-intent operations - the steps that map to real workflow tasks? Topic Center provides the design environment to define that mapping; the API Catalog provides the deployment path into Agentforce. For most organisations with an existing integration estate, the first agent-ready APIs can be live within days, not months.

The business case is already being proven in production. At Block - co-developer of the MCP standard - employees using MCP-connected AI agents report 50-to-75%-time savings on common tasks, with work that previously took days completing in hours. Default MCP servers connect to Snowflake, GitHub, Jira, Slack, and internal APIs, turning the agent into a single interface for workflows that previously required switching between six or more tools. Gartner forecasts that 40% of enterprise applications will include embedded AI agents by the end of 2026 - meaning the window for building an agent-ready API estate ahead of the curve is narrowing.

That is the practical outcome of agent-ready APIs: not a new system, but a new interface to the systems you already have - one that an AI agent can navigate, reason over, and act within, without a developer standing between every request and every response.

The brilliant new hire is waiting. The keys to the building are already in your API estate. Topics and MCP are how you give them out.

Want to understand how your existing API estate can become agent-ready with MCP and Topics? Talk to our team about a practical starting point.