Terraform MCP Server Enables AI Assistants to Interact with Terraform Infrastructure
HashiCorp's open-source Terraform MCP Server reaches GA, giving agents a sanctioned path to integrate with Terraform Registry APIs.
9 articles · 4 categories
What happened in AI — Jun 13, 2026
Saturday, Jun 13, 2026
In 30 seconds
- HashiCorp's Terraform MCP Server hits general availability — agents can now drive Terraform Registry APIs directly.
- Google moves WebMCP into Chrome 149 origin trials, letting sites expose tools to in-browser agents.
- Agent memory goes local-first: Cortex (encrypted, MCP) lands alongside a well-read survey of Letta, Mem0, Graphiti and Cognee.
- OpenDevOps debuts an open-source agent that investigates AWS/Azure incidents.
- Simon Willison surfaces a US export-control directive suspending access to Fable 5 and Mythos 5 on national-security grounds.
Saturday was an MCP day. The protocol that started life as a way to wire models to tools is steadily becoming the substrate of the agentic web: HashiCorp shipped its Terraform MCP Server to general availability, letting agents talk to the Terraform Registry directly, while Google pushed WebMCP — a proposal to let any website expose its own tools to in-browser agents — into Chrome 149 origin trials. A scrappier take, WebCLI, reframes the browser itself as 'just another agent skill.' The through-line: infrastructure and the open web are both growing first-class hooks for autonomous agents to act, not just read.
The other thread was memory and loyalty. A flurry of local-first projects argued that an agent's state shouldn't live in someone else's cloud — Cortex offers encrypted, local memory over MCP, a widely-shared explainer mapped the agent-memory landscape (Letta, Mem0, Graphiti, Cognee), and the draft Mutualism Accord tried to codify what a 'user-loyal' agent even means. Add Paca (a lightweight Jira alternative where humans and agents are equal teammates) and OpenDevOps (an agent that investigates AWS/Azure incidents) and you can see the same instinct from two directions: give agents durable context, but keep the human in control of it.
And then the kicker that overshadowed all of it for many readers: Simon Willison flagged a US government export-control directive ordering a suspension of access to Fable 5 and Mythos 5 on national-security grounds — a reminder that, however fast the tooling layer moves, the ground under the largest models can shift by decree.
MCP & the Agentic Web 3 items
MCP kept extending its reach today — into infrastructure tooling, the browser, and the open web — as the connective tissue for agents that act rather than just answer.
WebMCP Standard Proposal for Agentic Web Actuation Now Available in Chrome (Origin Trials)
Google's WebMCP enters Chrome 149 origin trials, letting sites expose JS functions and HTML forms as tools that in-browser agents can call.
Show HN: WebCLI – make the web browser just another agent skill
A Show HN project that reframes the browser as a callable agent skill, so navigating and acting on the web becomes another tool in the agent's belt.
Agent Memory & Local-First 3 items
A cluster of projects and writing converged on the same idea: agents need durable memory, and that memory should be local, encrypted, and loyal to the user.
Show HN: Cortex – local-first encrypted memory for AI agents (Rust, MCP)
A Rust, MCP-native memory layer that keeps agent state local and encrypted instead of in a vendor cloud.
Agent Memory Systems and Knowledge Graphs: Letta, Mem0, Graphiti, and Cognee
A practical survey of the agent-memory landscape, comparing how Letta, Mem0, Graphiti and Cognee approach persistent state and knowledge graphs.
The Mutualism Accord: a draft protocol for local-first, user-loyal AI agents
A draft protocol attempting to define what makes an agent genuinely 'user-loyal' and local-first, rather than aligned to a platform.
Human–AI Workflows & Ops 2 items
Two open-source projects put agents to work as teammates — one investigating cloud incidents, one planning sprints alongside humans.
OpenDevOps – An open-source AI agent that investigates AWS/Azure incidents
An open-source agent that triages and investigates AWS and Azure incidents, aiming to shorten the path from alert to root cause.
Show HN: Paca – Lightweight Jira alternative for human-AI collaboration
A Go-based, lightweight Jira alternative where humans and AI agents plan sprints and assign tasks to each other as equal teammates.
Safety & Policy 1 item
A reminder that the largest models sit on contested ground: government action, not just engineering, can pull access overnight.
Statement on the US government directive to suspend access to Fable 5 and Mythos 5
Simon Willison surfaces a US export-control directive that, citing national-security authorities, suspends access to Fable 5 and Mythos 5.