Articles

• How AI agents discover and select their own tools

  A practical model for runtime tool selection: describe the need, filter by what the agent can authenticate, check live health, call, then report.

• MCP reliability: why availability is more than an HTTP 200

  Liveness, capability, schema and data freshness are four different failure modes. How to probe MCP servers and APIs so agents are not misled.

• Why AI agents need a discovery and trust layer, not another directory

  Directories list servers for humans. Agents need machine-readable call contracts, live evidence and fallbacks. The difference, concretely.