Short answer: MCP security governs which tools an AI agent can discover and call, under which conditions, with which downstream credentials, and what data the tool may return.
Why an MCP connection is not a complete security policy
The Model Context Protocol standardizes how models discover and call tools. That solves interoperability, but enterprises still need to decide which agent can use each tool and what must happen before a sensitive action runs.
Transport authorization confirms that a caller may connect to a server. Tool governance goes further by controlling individual business actions, downstream credentials, request parameters, response data, human approval, operating limits, and audit evidence.
Controls required for governed MCP tools
- Policy-curated discovery. Return only the tools permitted for the authenticated agent and its assigned role.
- Per-call authorization. Evaluate every tool request rather than trusting access established at connection time.
- Separate downstream credentials. Do not pass an MCP client token through to an unrelated downstream API.
- Approval gates. Pause destructive, financial, or high-impact tools for an authorized person.
- Response controls. Remove sensitive data before tool output enters the model context.
- Complete audit records. Record the agent, tool, policy, approval, redaction, outcome, and time.
Where KeyRunner fits in an MCP architecture
KeyRunner acts as an MCP-compatible governed execution layer. An agent receives only the tools allowed by its named policy. When it calls a tool, KeyRunner checks the request before execution and resolves the separate downstream API credential inside the runtime.
This lets teams use MCP for interoperability while adding enterprise controls for tool access, credentials, approvals, rate and time limits, sensitive response data, and live audit evidence.
Enterprise checklist
- Filter tools by agent policy
- Validate every tool call
- Reject token passthrough
- Isolate downstream credentials
- Require approval for high-impact actions
- Export tool activity to audit systems
Frequently asked questions
What is MCP security and how do you govern MCP tools?
MCP security governs which tools an AI agent can discover and call, under which conditions, with which downstream credentials, and what data the tool may return.
Can KeyRunner govern tools used by Claude and ChatGPT?
Yes. KeyRunner exposes governed actions through MCP-compatible or API-based tool interfaces that can be used by compatible agent clients and frameworks.
Related KeyRunner guides
- How Do AI Agents Access APIs Without Exposing Credentials?
- How Do You Enforce Least Privilege for AI Agents?
- How Do You Audit API Calls Made by AI Agents?
- Agentic Security and AI Governance Guide
Explore the KeyRunner secure agent runtime, step through the live governance scenarios, or talk with the KeyRunner team.