APIs changed the world! But they’re showing their limits... Every integration still starts with studying formats, maintaining connectors, and debugging breakages after every version change.
The Model Context Protocol (MCP) is the start of something entirely different and it’s already happening NOW.
A new common language from Anthropic
MCP flips this model. Instead of sending data in a predefined JSON or XML structure, you simply ask a question or make a request in natural language. The connected system interprets it and responds with the relevant content using any internal tools it needs, without you needing to know how.
MCP was introduced by Anthropic, founded as a Public Benefit Corporation focused on AI safety and open collaboration, as an open standard for communication between AI models and external tools.
https://modelcontextprotocol.io/docs/getting-started/intro
No schema. No versioning. Just intent and response.
MCP is a conversational bridge between systems.
Every MCP-compatible app speaks the same protocol (something APIs never achieved, since each one has its own format and authentication style).
Agents can automatically discover available tools and prompts (while APIs require you to study each tool’s documentation before using it).
The Model Context Protocol (MCP) changes how systems talk. In ServiceNow, this aligns perfectly with AI Agent Fabric: allowing AI systems easily integrating together.
The problem APIs never solved
Tools like Zapier, Make, and IFTTT made automation accessible but did not remove the maintenance problem. Every new field, every changed endpoint, every version bump breaks your flow.
With MCP, that pain shifts away from you.
- No versioning debt: the MCP server handles schema evolution.
- Dynamic discovery: models ask “how do I do this?” and receive the latest valid method.
- Smaller interfaces: only relevant tools are exposed, keeping LLMs accurate and performant.
- Self-healing workflows: if a call fails, the model can introspect, re-prompt, and retry automatically.
- Natural language as configuration: integration finally moves from coding to conversation.
In other words, maintenance doesn’t disappear, it’s absorbed by the MCP layer.
For ServiceNow customers, this means fewer broken spokes, faster integrations, and a future where your AI Agents maintain themselves.
What MCP changes?
MCP is not a nicer API. It is a standard conversation layer! It’s “a multiprise universelle”, a universal socket, for agents to talk to tools.
This abstraction layer eliminates hard-coded schema dependencies: when Gmail or Notion change their internal APIs, the MCP server adapts, the client doesn’t.
It's now the MCP server to handle the details. When a provider changes something, the adapter absorbs it, the MCP layer adapts automatically and workflows keep running without modification.
How this fits ServiceNow
At ServiceNow, our mission is to be the AI experience layer for the enterprise. We already unify enterprise data through Zero-Copy, Stream Connect, and classic integration protocols. MCP will take this further, by allowing AI products and enterprise platforms to talk to each other naturally.
- No more API format investigation.
- No more custom spoke maintenance.
- No more Monday-morning integration errors.
This is not the future, it’s happening now. Early customers are already using it, and by the time you read this, it may be already be live (coming in the Zurich Patch 1 release - ServiceNow AI Agent as MCP client).
Example: from complex APIs to a simple MCP request
You want to compare flights and ask something like:
Find flights from Sydney to San Francisco on 10 December for 1 adult in economy. Keep it under AUD 1,500, with a maximum of one stop, and include only fares that allow one carry-on and one checked bag.
With a traditional API, you would need to:
- Learn the API structure: Convert every part of that request into the exact input format defined by the provider (for example, originLocationCode, destinationLocationCode, departureDate, adults, currencyCode, max, and includedCheckedBags), following their specific naming, structure, and data types
- Build and maintain the flow logic, handling dozens of actions
- Keep your integration updated every time the provider changes a field or version
With MCP, the request becomes a single line of intent:
- Find Sydney to San Francisco flights on 10 December for one adult in economy, under AUD 1500, with one carry-on, one checked bag, and no more than one stop.
- And that's it!
The MCP receiver translates that request into the internal logic it needs.
You get a ready-to-use, structured answer without having to know anything about the provider’s parameters or format.
Future is here!
APIs won’t disappear overnight, but the Model Context Protocol is showing what’s next: a universal language for systems to understand and collaborate.
It means that companies can shift their focus from run (executing repetitive maintenance) to build (designing powerful new automations)! And at ServiceNow, we’re ready to make that language enterprise-grade.
References:
- ServiceNow AI Agent Fabric overview
- Introducing AI Agent Fabric, MCP and A2A (Community)
- Underscore: Pourquoi l’automatisation par IA vient de franchir un cap ? (YouTube)
Published by Sylvain Hauser
From his internship at ServiceNow in 2007 to his certification as a Technical Architect (CTA), Sylvain has spent most of his professional life leading innovation with ServiceNow.
