MCP Security
Every MCP tool call, checked and recorded
Ferentin puts one MCP gateway between your agents and the servers they call. A server no policy grants is reachable by nobody, and every call is authorized, inspected and audited.
The problem
MCP hands agents real access
An MCP server gives an agent tools that read tickets, open pull requests and move data. Connected directly, each agent holds its own token for each server, and nobody sees the calls it makes or the results that come back. A tool description or result can also carry instructions to the model that no person ever reads.
How it works
How it works
- 1
Register servers and vet their tools
Add each server from the catalog, by URL or from a server card, and allow only the tools you need. Invisible characters in a tool’s name or description are shown as escape codes with a warning badge, never hidden. Ferentin also scans catalog metadata for prompt injection, typosquatting and planted credentials.
- 2
Grant tools by policy
A registered server no policy grants is reachable by nobody. An MCP policy names the servers, the tools on each, who it applies to and whether it allows, denies or holds a call for an approver. Start in Report only to see what a policy would refuse before it decides.
- 3
Inspect what comes back
Tool results pass through data protection, content sanitization and MCP Guard before they reach the model. Guard scores each result for prompt injection, jailbreaks and tool poisoning. It can log the detection, block the result or ask the user to approve it.
- 4
See every server and tool in use
MCP analytics shows which servers and tools are called, how often and with what error rate, for remote and local servers alike. Each row opens the calls behind it in Activity, with the user, the agent, the tool and the policy decision.
Works with your stack
Works with your stack
The AI clients your teams already use connect to one Ferentin MCP endpoint. The servers behind it come from the catalog or your own URLs.
Claude.ai
Claude Code
Cursor
ChatGPT
Visual Studio Code
Atlassian
GitHub
Salesforce
Box
Slack
Put a gateway between your agents and your MCP servers so every connection runs through one place you control. Register each server, allow only the tools it needs and grant access by policy, so a server no policy names is reachable by nobody. Inspect tool arguments and results for sensitive data and prompt injection. Record every call. On laptops, route local MCP servers through a governed relay so they are covered too.
Go deeper
Learn more
- MCP server access: direct vs gateway-brokeredWhat changes for identity, credentials and audit when agents reach MCP servers through a gateway.
- Glossary: Tool PoisoningHow hidden instructions in tool descriptions reach the model, and how to catch them.
- Apple is turning Mac apps into MCP serversSafari and Xcode ship local MCP servers that no gateway sees. What governing them takes.
Related
- Enterprise-Managed AuthorizationYour IdP approves each agent connection. Ferentin authorizes every call that follows.
- AI & MCP GatewayOne governed path to every model and MCP server, managed or in your cloud.
- Agent Runtime SecurityEvaluate each agent action in context. Stop toxic flows before they run.
- AI Data Loss PreventionRedact or block secrets and personal data in prompts, tool calls and results.
See it on your own stack
Walk through it with our team on your identity provider, your models and your MCP servers. Or start free today.