Model Context Protocol (MCP) Server Directory

Microsoft is fundamentally changing how AI agents operate in production by introducing strict MCP (Model Context Protocol) allowlists for GitHub Enterprise.

Package AI agent capabilities into portable, vendor-neutral Agent Plugins that bundle Skills with MCP servers, then automate schema validation, multi-client smoke tests, and catalog publishing.

Architect a stateless MCP server on Cloudflare Workers that survives restarts between rounds of a human approval gate using MRTR requestState, the tasks extension, and RFC 9207 OAuth verification.

Operate OpenAI's GPT-5.6 programmatic tool calling in production: let Luna, Terra, or Sol author typed JavaScript that orchestrates allowlisted tools inside the OpenAI sandbox, with approval gates, step budgets, and OpenTelemetry tracing.

Google's Agentic Resource Discovery (ARD) protocol, built under the Linux Foundation AAIF, lets an agent ask what is available before anything runs. This guide builds a stateless ARD-style MCP discovery server that indexes plugins, MCP servers, and skills and answers query-by-task searches with ranked, cacheable results.

Agent Plugins 1.0.0 is the Linux Foundation AAIF's vendor-neutral bundle format from Amazon, Cursor, Microsoft, OpenAI, and Vercel. This guide builds a stateless FastMCP installer that fetches, validates, maps, and installs plugin bundles into Claude Desktop and Cursor - and uninstalls or updates them on demand.

On August 6, 2026, the Linux Foundation's Agentic AI Foundation shipped Agent Plugins 1.0.0 — the first truly vendor-neutral way to package Agent Skills and MCP servers into one portable, write-once-run-anywhere directory. Here is the full technical breakdown and what it means for your build.

Mira Murati's Thinking Machines Lab shipped its first model: Inkling, an Apache-2.0 Mixture-of-Experts built deliberately as a fine-tuning base, not a frontier competitor. Here is the MoE fine-tuning strategy, quantization reality, and the fine-tune vs RAG vs prompt decision framework.

GPT-5.6, GA on July 9 2026 as Luna, Terra and Sol, added programmatic tool calling: the model writes JavaScript that orchestrates your tools inside OpenAI's sandbox instead of round-tripping every call through the conversation. Here is a worked example, the security model, and the budgets that make it safe.

For the first time in AI history, enterprise spending on AI inference has officially surpassed investments in model training, signaling a massive industry shift toward deployment and ROI.

Transition from MCP 2025 stateful sessions to the new 2026 stateless architecture with our comprehensive MCP SDK v2.0 migration guide.

Implement centralized policy enforcement for AI agents with a GitHub Copilot Enterprise MCP allowlist server managing allowed and denied tools.