This Week in Agentic AI: September 14–21, 2026
MCP integrations reached messaging and smart-home workflows, while Claude Code Projects introduced shared context for multiple agents. The week also highlighted shared memory, credential handling, and the controls needed when agents connect to everyday services.
MCP reaches messaging and the home
Meta introduced a WhatsApp Business MCP server for setup, templates, testing, and troubleshooting. Google announced early access to a Google Home MCP server, and Simon Willison discussed MCP’s role in authentication, connection interfaces, and audit logging.
Shared context and credentials
Claude Code relaunched Projects with shared memory, goals, files, and coordinated threads. The llm-keys-ui plugin offered a local interface for managing keys used by coding agents.
Household coordination and oversight
Google repositioned CC around family schedules, tasks, and household coordination. Instinct and Muse added calling capabilities, while coverage of agent-monitoring tools and startups focused on responding to misbehavior.
Top stories this week
MCP was always a bad idea? Simon Willison responds
Simon Willison argues that while full terminal coding agents with unfettered internet can call APIs directly, MCP provides essential controls, authentication handling, connect UI, and audit logging for less 'YOLO' agent setups.
Why it matters for builders
For developers building agents that connect to external services, MCP lets you restrict access, handle auth without exposing API keys to the agent, offer user-facing connect flows, and maintain audit logs—practical building blocks that raw API calls don't provide.
Claude Code relaunches Projects to manage multiple AI agents in the cloud
Anthropic has reintroduced Projects in Claude Code, allowing users to run multiple AI agents with shared memory, goals, and a common file library. Each project contains parallel threads managed by a coordinator agent.
Why it matters for builders
Developers can now orchestrate multiple Claude Code agents in a cloud workspace with shared context, reducing the overhead of managing state across parallel coding tasks without self-hosting coordination.
Meta lets AI agents handle WhatsApp Business setup via new MCP server
Meta introduced a WhatsApp Business MCP server that allows AI coding agents like Claude, Cursor, Codex, and ChatGPT to automate setup, messaging templates, testing, and troubleshooting.
Why it matters for builders
Developers integrating WhatsApp Business can offload repetitive API configuration and testing to coding assistants, reducing onboarding friction and ongoing maintenance work.
Your AI agents can now control your Google Home devices
Google is launching early access to an MCP server for Google Home that lets AI assistants like Claude and ChatGPT control smart home devices, review camera summaries, and access activity using natural language.
Why it matters for builders
Developers can now connect Google Home devices to agent workflows via MCP, enabling voice or chat-driven home automation without building custom smart home APIs. This opens up new use cases for home management agents and multi-device orchestration.
llm-keys-ui 0.1
Simon Willison released llm-keys-ui 0.1, a plugin that provides a local web interface for storing API keys used by coding agents. It lets Codex Remote retrieve keys via CLI commands like `llm keys get anthropic` instead of pasting them into chat sessions.
Why it matters for builders
For developers running coding agents on remote machines, this plugin avoids pasting API keys into agent chat sessions by offering a web UI accessible over local network or Tailscale IPs, with keys retrievable via shell commands. It's a practical pattern for injecting secrets into agent workflows without manual copy-paste.