Intelligent automation startup Celigo Inc. today announced its Model Context Protocol Management platform, which will let information technology and security teams govern how artificial intelligence assistants and agents act on enterprise systems.
Model Context Protocol, or MCP, has become a standard for connecting AI assistants and agents to business systems and services. Many enterprise services currently host MCP servers to provide access to business systems, and these servers expose tools that let agents interact with basic and complex controls.
Anthropic PBC introduced MCP in 2024 as an open-source standard, often described as a USB-C port for AI. It acts like a single, universal access point for AI models and agents to plug into databases, local files, search engines, business data and application programming interfaces to access resources.
Examples include looking up customer information in a customer management system, checking invoices in an enterprise resource planning system, updating an order status in a parcel tracking system or working through a ticket on a service desk. The sort of work that used to fall to integration specialists can now be completed by AI agents such as Claude, ChatGPT, Copilot and Cursor autonomously.
This opens up numerous potential security risks and access issues if these agents have access they shouldn’t or take actions outside of their role. So, it’s important to carefully limit their activity and guide them on where they operate. Although AI agents can act autonomously and provide a great deal of power, they can also hallucinate – that is, make mistakes as they work with little supervision – and they can be hijacked by malicious actors.
“AI assistants and agents are a new class of actors in the enterprise and the old model of shared credentials and broad access was never designed for them,” said Chief Product Officer Matt Graney. “Enterprises need to know who is behind every AI request, what it is authorized to do and what actions it has taken.”
Celigo MCP Management gives IT a single governed path to lock down agents and keep them within their respective domains. Graney added that every call traces back to a real person, access is approved one tool at a time and the platform runs the tools and controls them, so teams can see which system the AI has touched and maintain full transparency.
This centralized control enables broad, granular management of AI assistants and agents across the enterprise.
The company said this management interface connects MCP servers to grounded company knowledge to more than 1,000 apps built for company jobs. It lets teams control who uses which tools with an MCP Gateway and discover every MCP server through a registry.
“Celigo’s MCP was the missing piece. It’s the layer in the middle connecting to all our third-party tools in the cloud. Now we get the start, the middle, and the end in one dashboard,” said Nidecker SA Head of IT Antoine Millet.