MCP Articles
14 articles across AI Catchup's news, guides, tutorials, and comparisons.
All MCP articles
ChatGPT Sites Can Turn MCP Tools Into Plugins and Shareable Work
ChatGPT Sites can host an MCP server whose tools are installed as a plugin in ChatGPT or Codex. Publishing the Site creates the plugin, but access and connected-account authorization still apply. Shareable profiles can showcase Sites and plugins, with privacy, workspace, device, plan, and rollout limits to check.
Claude Opens a Plugin Directory Submission Portal for MCP and Skills
Anthropic launched a Claude directory submission portal on September 25, 2026. Developers on paid Claude plans can submit a remote MCP connector or a GitHub-hosted plugin bundle that packages MCP servers and skills, follow validation and safety review, publish approved versions, and see listing and usage analytics.
OpenAI Opens Its Codex-Powered Agents API to All Developers in Public Beta
OpenAI's Agents API is now in public beta for all developers. The managed service uses the Codex harness to run long-lived cloud agents, coordinate parallel subagents, compact context across long sessions, and work in OpenAI-hosted, VPC, or partner-provided sandboxes.
Claude Makes Enterprise-Managed Auth for MCP Connectors Generally Available
Anthropic says enterprise-managed auth for MCP connectors is now generally available. Claude Team and Enterprise admins can centralize authorization through their identity provider, while users receive connector access automatically without individual OAuth consent flows.
Agent Plugins: An Open Standard for Skills and MCP
OpenAI Developers introduced Agent Plugins, an open, vendor-neutral package format for sharing Agent Skills and MCP server configurations across compatible agent clients.
Cursor Says Cloud Agents Use 20-30% Fewer Tokens
Cursor said on August 3, 2026 that its Cloud Agents are 20-30% more token efficient, and 80% more efficient on runs with computer use, crediting better MCP, skills, and computer-use handling. A September 23 agent-harness update later reported a further 7% cut in token costs without lower agent quality.
MCP 2026-07-28 Moves to a Stateless Core as Claude Rolls Out Support
Anthropic says the MCP 2026-07-28 specification moves the Model Context Protocol from a bidirectional stateful design to a request/response model, adds versioned extensions for MCP Apps and Tasks, and aligns authorization with production OAuth 2.0 and OIDC deployments.
Claude Code artifacts can now call MCP connectors for live, viewer-specific data
Claude Code artifacts can call MCP connectors each time someone views them, so a published page shows current data rather than a session snapshot. Available on Pro, Max, Team, and Enterprise plans with Claude Code v2.1.209 or later. A connector-backed page can now go to a public link, but its live sections stay empty for signed-out or outside viewers.
OpenAI's Secure MCP Tunnel: Connect Private MCP Servers Over Outbound-Only HTTPS
OpenAI's Secure MCP Tunnel connects private and on-prem MCP servers to ChatGPT, Codex, and the Responses API without opening inbound firewall ports. You need a tunnel_id, an API key with Tunnels Read plus Use, and -- separately -- ChatGPT developer-mode access. That permission split is the most common setup blocker.
Claude Managed Agents Add Self-Hosted Sandboxes (Public Beta) and MCP Tunnels (Research Preview)
Anthropic says Claude Managed Agents can now run tool execution in a sandbox you control (public beta) and connect to private MCP servers via MCP tunnels (research preview). The update targets enterprise security requirements by keeping execution and private services within an organization's perimeter.
Claude Code 2.1.142: `claude agents` Gains Session Flags, Fast Mode Defaults to Opus 4.7, MCP Tool Timeout Honored
Claude Code 2.1.142 (May 14, 2026) added eight dispatch flags to `claude agents`, moved fast mode's default from Opus 4.6 to Opus 4.7, and made `MCP_TOOL_TIMEOUT` lift the 60-second cap on remote MCP servers. The flags and timeout fix stand; fast mode now defaults to Opus 5.5 (v2.1.280 and later) at $8/$40 per MTok.
Claude Code 2.1.133: `worktree.baseRef` Default Returns to `origin/<default>`, MCP OAuth Proxy Honored Across the Whole Flow
Anthropic shipped Claude Code 2.1.133 on May 7, 2026. The headline is a worktree-base behavior change: a new `worktree.baseRef` setting (`fresh` | `head`) defaults to `fresh`, which moves `EnterWorktree`'s base back to `origin/<default>` after three days of branching from local `HEAD` (since 2.1.128 on May 4). The release also routes `HTTP(S)_PROXY` / `NO_PROXY` / mTLS through the entire MCP OAuth flow (discovery, dynamic client registration, token exchange, refresh), exposes effort level to hooks via `$CLAUDE_EFFORT`, adds Linux sandbox path overrides, and fixes a refresh-token race that was 401-ing parallel sessions.
AI Tools Landscape: What Changed in Early 2026
Three shifts defined AI tooling in early 2026: MCP settled as the cross-tool standard, coding assistants grew past autocomplete into multi-file workflow partners, and narrowly autonomous agents reached production on supervised tasks. MCP's tipping point came in June 2025, when Visual Studio Code made MCP support generally available.
How to Build Your Own MCP Server: A Beginner's Guide
To build an MCP server in TypeScript, install the v2 SDK package @modelcontextprotocol/server with Zod v4, register each tool with server.registerTool and a Zod input schema, serve it over stdio, then add it to Claude Code with claude mcp add or a .mcp.json entry. A one-tool server is about 50 lines.