MCP Servers

MCP Servers extend your agents with external tools and capabilities through the Model Context Protocol. They provide a structured way to connect APIs, services, and internal systems without hard-coding each integration in agent prompts.

Core Capabilities

• Discover toolsets exposed by connected MCP servers.
• Select specific servers per agent or per agentspec.
• Observe server status and tool availability at runtime.
• Route tool calls through protocol-safe interfaces.

Common Use Cases

• Research and web retrieval.
• Enterprise API orchestration.
• Internal knowledge and workflow integration.
• Tool-rich agents with explicit approval or policy gates.

Operational Pattern

1. Register available MCP servers in your environment.
2. Select required servers in your agent config/spec.
3. Launch agent and verify MCP status indicators.
4. Execute tool calls through chat or workflow triggers.
5. Monitor usage, failures, and latency.

Best Practices

• Start with a minimal server set per agent.
• Expose only needed tools for least-privilege behavior.
• Use guardrails and approvals for high-impact tool actions.
• Track tool health and aggregate status in monitoring views.

Next Steps

• Build runtime policies in Guardrails.
• Configure execution in Sandbox and Codemode.

See also

• Tool Approvals
• Monitoring
• Outputs