Bonfy's latest MCP Server feature marks a breakthrough in securing AI agents by enabling real-time data inspection during their reasoning process. This v2.0 update addresses critical data leakage risks in agent workflows, positioning Bonfy as the data-centric leader in AI security.
AI agents, lightweight software powered by LLMs like those from OpenAI or Microsoft, connect data sources, tools, and services to execute tasks autonomously. They introduce multiple leakage points: initial prompts, unauthorized data access, communications with external MCP servers, and final outputs like emails or files. Unlike static Copilot interactions, agents perform actions across systems, amplifying risks in enterprise environments where 99% run on hosted platforms like Microsoft Power Platform or Salesforce Agentforce.
Traditional tools focus on configuration management—cataloging agents and their tools, but ignore data flows. Endpoint DLP fails for cloud-hosted agents, and browser extensions miss backend operations. Bonfy shifts to data-layer protection, inspecting content regardless of agent location or configuration.
Bonfy layers security across the agent lifecycle. Input control uses contextual classification to limit grounding data, e.g., SharePoint access by sensitivity labels. Output control scans final products before outbound channels. The new MCP Server provides data-in-use inspection: agents call Bonfy's server mid-reasoning to verify content safety.
In practice, prompts/agent-instructions instruct agents: "Verify with Bonfy for PII/other risks before summarizing." The agent calls Bonfy's MCP (REST APIs supported at launch), receives risk ratings, and adjusts—proactive, not reactive. This unified platform uses the same engine and policies across layers, simplifying deployment.
Consider an agent summarizing customer emails: it pulls data, calls Bonfy's MCP Server ("Is this safe?"), gets entity-aware analysis via Bonfy's Knowledge Graph, and proceeds or halts based on results.
Competitors chase configs; Bonfy secures flowing data, like software abuse prevention. V2.0's low-latency architecture enables this real-time capability.
Proactive verification reduces incidents, cuts false positives with entity awareness (people, customers behind data). Complements IAM and posture tools in the "three-legged stool" of agent security: data, config, identity. Greenfield orgs get an all-in-one foundation; hybrids fill AI gaps.
As agents proliferate (Copilot Studio, Claude, OpenClaw), Bonfy ensures safe scaling without friction. Early adopters gain compliance, trust, and efficiency.
To learn more, click here.