Phase 13
Tools and Protocols
The interfaces between AI and the real world.
Lessons (31)
- 01The Tool Interface — Why Agents Need Structured I/O
- 02Function Calling Deep Dive — OpenAI, Anthropic, Gemini
- 03Parallel Tool Calls and Streaming with Tools
- 04Structured Output — JSON Schema, Pydantic, Zod, Constrained Decoding
- 05Tool Schema Design — Naming, Descriptions, Parameter Constraints
- 06MCP Fundamentals: Stateless Requests and JSON-RPC
- 07Building an MCP Server: Stateless Python and TypeScript
- 08Building an MCP Client: Discovery, Routing, and Dual-Era Fallback
- 09MCP Transports: stdio and Stateless Streamable HTTP
- 10MCP Resources and Prompts: Addressable Context for Stateless Servers
- 11MCP Model Input: Sampling Migration and Stateless MRTR
- 12Explicit Scope and Stateless Elicitation
- 13MCP Tasks Extension: Durable Work on a Stateless Core
- 14MCP Apps on the Stateless Protocol
- 15MCP Security: Poisoned Metadata, Routing, and MRTR State
- 16MCP Authorization: CIMD, Issuer Binding, PKCE, and Step-Up
- 17Stateless MCP Gateways and Registry Admission
- 18MCP Auth in Production: Issuer-Bound Enrollment and Tokens
- 19A2A — Agent-to-Agent Protocol
- 20OpenTelemetry GenAI — Tracing Tool Calls End-to-End
- 21LLM Routing Layer — LiteLLM, OpenRouter, Portkey
- 22Agent Skills: Portable Contract and Runtime Boundary
- 23Capstone: Stateless Tool Ecosystem
- 24Skill Discovery and Progressive Disclosure
- 25Skill Invocation and Routing
- 26Skill Permissions, Sandboxes, and Trust
- 27Skill Evals, Packaging, and Portability
- 28MCP Tool Contracts and Content
- 29MCP Reliability, Cancellation, and Flow Control
- 30MCP Registry Supply Chain: Admission, Drift, and Rollback
- 31MCP Conformance Engineering: Versioning, Evidence, and Operations