MCP Bastion is a security proxy for MCP (Model Context Protocol) servers, used to add a layer of security control between AI agents and MCP tools.
Main capabilities:
- Tool-definition pinning: locks tool definitions so that changes to definitions can be detected, reducing the risk of tools being tampered with.
- Tool-poisoning detection: identifies potentially poisoned content in tool descriptions or definitions.
- Exfiltration detection: detects possible data exfiltration behavior.
- Injection blocking: blocks prompt injection attacks.
- Secret redaction: redacts sensitive information.
- Audit trails: records calls and security events for traceability.
Usage and limitations: the current material does not provide specific installation, configuration, or run commands, nor does it yet explain supported platforms, deployment methods, pricing models, or effectiveness metrics; these should be based on the project repository's subsequent documentation.
