Model Context Protocol (MCP) Server Directory

Koboldcpp v1.120 ships DirectIO loading and Qwen 3.8-Flash-Next support. Build a FastMCP server that lets Claude Desktop discover, load, switch, and query local open-weight models through the MCP protocol.

Tencent just dropped Hy4-preview: a 770B-parameter Mixture-of-Experts model under Apache 2.0 with 1M-token context. Here is how to deploy it locally with vLLM 0.28.0 and orchestrate a multi-agent workflow that routes tasks across 256 expert shards.

Faro AI powers structured clinical data for 6 of the top 10 pharma companies. Build a FastMCP server that exposes trial-protocol search, patient-cohort matching, and regulatory dossier generation as MCP tools for healthcare agents.

Tencent's Hy4-preview 770B model is available under Apache 2.0. Build a FastMCP server that exposes local Hy4 inference as an MCP tool for Claude Desktop and Cursor IDE.

Artiforge is not just another MCP server — it is a complete AI development toolkit. Build a FastMCP Python server that exposes code generation, test writing, refactoring, and documentation tools to Claude Desktop agents.

Sprinklr's Summer 2026 release quietly added MCP beta access, joining HubSpot and Salesloft in the enterprise martech-to-agent pipeline movement. Here is what this means for AI-powered customer experience.

Sprinklr's Summer '26 release added MCP beta access. Build a production FastMCP TypeScript server that exposes enterprise customer data, campaign metrics, and LLM-powered insights to Claude Desktop and Copilot agents.

Context7 has become the #1 ranked MCP server in 2026 for autonomous code generation. Build a FastMCP TypeScript server that provides real-time library documentation to AI agents, eliminating hallucinated APIs.

AI agents need instant access to web analytics. This FastMCP server exposes Vercel Analytics and Edge Config data to Claude and Cursor, querying 50 million page views in under 3 seconds.

AI agents need project management access. This FastMCP server exposes Linear issue tracking to Claude and Cursor, autonomously triaging 500 issues per hour with 94% accuracy.

AI agents need live database access. This FastMCP server streams Supabase Realtime changes, invokes Edge Functions, and queries pgvector — giving Claude and Cursor instant access to live application data.

Microsoft Agent Framework 1.0 shipped in April 2026, merging AutoGen and Semantic Kernel into a single production-ready platform. This pipeline deploys three specialized review agents (security, performance, style) orchestrated by MAF 1.0 that automatically review GitHub PRs and post consolidated feedback.