#agents

Anthropic's own four-phase guide to building an MCP server — research, implement, review, then generate ten evaluations that test whether a model can actually use it.

Scaffolds and validates an Agent Skill — the packaging format that puts your conventions, gotchas and idiomatic examples inside the agent your users already work in.

Probes robots.txt, llms.txt and llms-full.txt, scores AI-readiness against a structured checklist, and needs no API key.

The 2026-07-28 spec drops MCP's session handshake, hardens auth, and ships day-one across four SDKs with six hyperscalers behind it — the protocol crossed from experiment to backed standard, which resets both the positioning window and the migration bill.

In two weeks at least five teams shipped the same primitive — a policy gate between an AI agent and an irreversible action — and independent research is now naming it. The category is forming; the positioning window is open.

Sinch launched Agent Tools on 2026-08-04: VS Code/JetBrains/Open VSX extensions, an MCP server exposing live Sinch API definitions, a Simulator Mode that tests integrations before an account exists, and "Sinch Skills" — structured implementation guidance for AI coding assistants including Claude Code, Cursor, GitHub Copilot and ChatGPT Desktop. It joins a growing roster of vendors making their platforms "agent-safe by design" rather than just documented for humans.

Y Combinator-backed Armature launched on Show HN with product analytics that wrap your MCP server to reconstruct agent sessions — user intent, agent reasoning, task success and failure — the way PostHog or Amplitude reconstruct a human click stream. It picked up 40 points, and it is not the first entrant: AgentCat (mcpcat.io) already sells the same "agent experience" category.

Tines launched 3B on Show HN with the sharpest positioning of the week — "you told everyone to use AI, now give them a secure place to do it" — selling governed visibility over AI-built workflows, dashboards and automations rather than gating who is allowed to build. It picked up 27 points and 2 comments, modest traction for a category that has been filling with near-identical guardrail entrants all month.

Recorded at Heavybit DevGuild, the episode traces how Unikraft went from an academic unikernel project to a cloud platform for efficient workloads, and covers why unikernels can make applications faster, cheaper and more secure. The framing worth noting for a devtools marketer is the demand story: AI agents needing disposable sandboxes is what turns a research-grade technology into a category with buyers.

Kastra shipped a runtime authorization layer that intercepts agent tool calls with identity, scope, guardrail and audit checks at sub-millisecond allow/deny, and it covers Claude Code, Cursor and Codex rather than integrating with one harness. Traction is modest — 13 points and 5 comments — on a Show HN posted around 2026-07-10 that only cleared this site's sweep on 07-28.

The 2026-07-28 Model Context Protocol spec drops `initialize` and session IDs so any server instance can answer any request behind a plain load balancer, graduates Tasks and MCP Apps into a versioned extensions framework, and hardens auth with RFC 9207 issuer validation in place of Dynamic Client Registration. All four Tier-1 SDKs shipped support on publication day, with AWS, Cloudflare, Figma, Google Cloud, Microsoft and Netlify publicly committed.