After Google Put REST on API Gateway MCP, Why Is Discovery Still JSON? OpenAPI 3.x to tools/list
The gateway plays MCP server. Discovery is still tools/list JSON. Lock the catalog and Diff it against your OpenAPI.
JSON tips, comparisons, and best practices to help you understand and work with JSON data more deeply.
The gateway plays MCP server. Discovery is still tools/list JSON. Lock the catalog and Diff it against your OpenAPI.
Sonnet 5.5 keeps the sticker and rejects disabled. For JSON extraction, use adaptive thinking plus Schema — not between_tools.
Opus 5.5 rejects forced tool_choice. Stabilize JSON with auto + strict + Schema before you change the model string.
A Skill on MCP is still discovered as JSON. SEP-2640 is Accepted, not Final — check the index before you load SKILL.md.
Hosting the harness is not hosting the contract. After the Agents API, JSON Schema is the layer you still hold.
Looks like JSON still throws. Strip fences first, prefer Structured Output, and do not start with a regex repair.
A jailbroken chatbot says the wrong sentence. A jailbroken agent fires a tool call. Tighten JSON Schema before you tune the prompt.
Tools are the call contract; MCP is the cross-process socket; Skills are how-to playbooks; Subagents isolate context. Mix the layers and you debug the wrong one.
What can stay on device, stays. What must go to the cloud uses PCC — not a normal LLM API. Leaks usually start in the JSON fields you put on an App Action.
Not a chatbot skin. 12GB RAM, Siri AI, and App Actions turn the iPhone 18 Pro lineup into a system-level AI agent — with JSON Schema as the contract.
Structured Output is not the prompt “please return JSON.” GPT, Gemini, and Claude now block illegal tokens with a JSON Schema so a program can parse the final reply.
MCP is not another Function Calling. It is JSON-RPC between Host and tool processes; the model still uses Tool Calling, with JSON Schema as the contract.
1M tokens is not “a million Chinese characters.” GPT-5.5, Claude 4.6/4.8, and Gemini 3.7: advertised window, effective retrieval, and the bill for stuffing large JSON.
Ternus took over on 1 September. Siri AI, Apple Intelligence, the $1B Gemini license, and what an engineer-CEO can change before 9 September.
Gemini 3.8 Flash is still not GA. September 2 launch rumors, the release window, and an AI coding forecast vs 3.7 Flash.
Gemini 3.8 Flash is not GA yet. August 2026 leaks, launch-window math, performance outlook vs 3.7, and API JSON compatibility.
OpenAI vs Gemini Structured Output: 2026 JSON Schema API fields, strict subsets, code samples, and dual-provider migration.
Why Tool Calling needs JSON Schema, common parameter/type errors in arguments, and validation methods from ajv to strict mode and error feedback.
Is JSON Schema becoming the default Agent contract? Tool Calling, Structured Output, MCP inputSchema, and where OpenAPI still fits.
Gemini 3.7 Flash intro pricing, gemini-3.7-flash API, Structured Output, Function Calling, and migration from 3.6 Flash.
From prompt-only JSON to Structured Outputs — JSON Schema constraints, OpenAI / Gemini / Anthropic comparison, validation pipeline, and Tool Calling.
From prompt-only JSON to responseSchema / JSON Schema — Gemini structured output, Python and REST examples, and how to validate before you ship.
Trace every JSON hop in an Agent call: Schema, Function Calling / Tool Calling, MCP JSON-RPC, and how validation failures flow back.
What changed in MCP 2026, whether your MCP Server needs code changes, plus migration steps, a compatibility checklist, and JSON Schema validation tips.
Rankings and hands-on reviews of the most worthwhile MCP Servers in 2026 — by stability, security, ecosystem, and real-world scenarios, with category picks, setup tips, and pitfalls to avoid.
From plain-text chat to executable agents — how JSON Schema, Function Calling, and MCP became AI Agent infrastructure, with a timeline, synergies, and practical choices.