MCP Security and API Isolation
Model Context Protocol is how agents reach your systems. Without hardening, agents may call tools and APIs they were never intended to reach.
Fortytwo help you assess and harden the MCP layer and agent-facing APIs, establishing authorization controls, execution isolation, and tool registry governance so agents can only call what they are explicitly permitted to call.
The Challenge
MCP is a common connector for agents.
Overprivileged tools and unauthenticated endpoints can turn it into an open door, and tool poisoning and prompt injection can be used through the MCP layer.
Every tool call is made by a non-human identity, and without scoping and clear boundaries, agents may access more things than they should.
The Solution
Agents that call only what they are permitted to call
Fortytwo helps you:
What We Deliver
Key Outcomes:
MCP servers in scope are inventoried, authenticated, and configured according to least-privilege principles. Agents are constrained to their explicitly authorized scope at the API layer, making tool poisoning and prompt injection risks introduced through the MCP layer more readily detected and mitigated. Agent API and tool-call activity is also logged and reviewable in the SIEM.
MCP Security Design
An assessment and hardening design for MCP servers, covering authentication, scoped authorization, and supply-chain risk.
API Authorization Model
A least-privilege permission model for agent-to-API binding, using OAuth 2.0 and managed identity where applicable.
Agent Isolation Architecture
Network segmentation, execution sandboxing, and a per-workload blast-radius containment approach.
Secure Tool Registry
A registry of approved MCP servers and versions, with tool poisoning and prompt injection detection.
MCP Security Runbook
An incident response playbook, with MCP tool-call logs integrated into SIEM for anomaly detection.
How It Works
We start with an assessment of your current environment, then
moving to and assessment of the MCP layer. After this, we design the authorization and isolation model, then support the implementation with the first tools onboarded, and new tools following the same pattern through the registry.
The result is a reusable, hardened platform rather than a one-off integration
Related Services
FAQ
We need access to Azure API Management, your Azure AI Foundry and MCP servers, and Microsoft Entra for managed identity. We also need owners nominated for the tools in scope.
In most cases, no. Many teams use existing servers like the Azure MCP Server. We harden what you have and build custom servers only where you need them.
Both are in scope. We build a secure tool registry, support detection of poisoning and injection at the MCP layer, and define containment by workload.
API Management acts as the enforcement gateway. MCP is the protocol behind it. If you already run APIM, much of the pattern will be familiar.
Talk to Us
Get in touch if you want to discuss your challenges or questions.
Harri Jaakkonen
Principal Security Engineer
oi.owtytrofobfsctd-52edb5@nenokkaaj.irrah
The protocol is the control point, and you should treat it accordingly.
