AISafetyHot

What AISafetyHot does

AISafetyHot gives MCP-compatible agents a query interface for the AI Safety HOT service. The service curates AI safety news, papers, incidents, and related developments covering areas such as jailbreaks, prompt injection, agent security, defenses, alignment, evaluations, and governance policy.

Its public data includes recent selected news, ranked hot topics, event coverage, daily updates, and daily paper recommendations. Agents can use the service to prepare briefings, investigate a developing incident, monitor an organization or product, or find research in a particular safety area. Results link back to information collected from external sources, so the service works as a research and discovery layer rather than an authoritative primary source.

How it works

The MCP endpoint is available at https://aisafetyhot.com/api/mcp using Streamable HTTP. An MCP client sends tool calls to the endpoint and receives structured results that the agent can summarize or use for follow-up queries. No API key is required for the public access tier.

The repository also includes four agent skills: aisafety-hot, aisafety-brief, aisafety-paper-scout, and aisafety-watch. These skills provide task-oriented instructions for asking about hot topics, producing daily safety briefings, scouting papers, and watching a company or product.

Setup and configuration

Configure an MCP client with a Streamable HTTP server at:

https://aisafetyhot.com/api/mcp

Claude Code can install the MCP server and skills through the aisafetyhot plugin. Codex can register the endpoint with codex mcp add, while skills can be added with npx skills add wuyoscar/AISafetyHot. Cursor and other compatible agents can add the same URL in their MCP settings. The repository also provides install, update, and removal instructions for supported workflows.

A basic health check calls aisafetyhot_get_hot_topics with a result limit. The service returns HTTP 429 when an IP exceeds its request allowance.

Tools and capabilities

Available examples include:

• aisafetyhot_get_hot_topics for ranked current topics, with a configurable limit shown in the usage example.
• aisafetyhot_get_story for examining the reports and timeline associated with a hot event.
• aisafetyhot_get_latest for recent selected updates, including topic-filtered queries such as anthropic.
• Daily paper discovery across AI safety research areas, with summaries, original publication details, links, and BibTeX support through the paper skill.

Limitations and notes

The public interface is read-only: it retrieves information but does not create or modify records. Anonymous access covers selected news from the last seven days, hot-topic events, daily reports, and daily papers. Older historical queries, complete paper interpretations, and weekly or monthly reports require read-only authorization.

The service limits each IP address to 60 requests per minute and 300 requests per day. Returned titles and summaries originate from external sources; agents should treat them as reference material rather than instructions to execute. Streamable HTTP support is required from the MCP client.