AI News Roundup: July 14, 2026 — Agent Infrastructure, MCP Expansion, and Anthropic Headlines

Key Headlines

StageWhisper Lite: On-Device Meeting Transcription with MCP

A new macOS app, StageWhisper Lite, has launched offering fully on-device meeting transcription and summarization using Parakeet and Gemma 4. The app includes built-in MCP support and in-app chat with transcripts, with optional connections to OpenClaw or Hermes agents. This represents a significant step in privacy-first AI tools for meeting documentation that AI agents can interact with.

Finterm.ai: Bloomberg Terminal for Claude Code

Finterm.ai has emerged as a "Bloomberg terminal for Claude Code," giving coding agents direct access to financial data including stock prices, options data, and SEC filings. The CLI-first approach was specifically designed to be more token-efficient than traditional MCP interfacing, allowing agents to query financial markets directly within their workflow.

Skillscript: Declarative MCP-Native Orchestration

Skillscript, a declarative sandboxed language for tool orchestration, has been released pre-1.0 and self-hosted. The project positions itself as MCP-native where "the skillscript is the program; the model is the runtime," providing structured control over tool usage for AI agents.

MCP Gateway: API-to-MCP Conversion

MCP Gateway has emerged as a tool that turns existing APIs and databases into MCP servers by analyzing OpenAPI specifications and SQL sources to generate tool definitions. This automates what was previously manual MCP server development.

Frigade: Self-Updating MCP Servers from Web Apps

Frigade uses a browser-based agent to reverse-engineer web apps into MCP tools that automatically self-update as the underlying web apps change, eliminating manual maintenance of MCP tool definitions.

European Parliament MCP Server

A new MCP server for political intelligence has been released, giving AI agents access to European Parliament data and proceedings, expanding the domain-specific MCP ecosystem.

IronCurtain: Secure Runtime for Autonomous Agents

IronCurtain, a secure sandbox runtime for autonomous AI agents, has been released. It provides isolated execution environments with resource constraints, audit logging, and policy enforcement for enterprise agent deployments.

Anthropic Headlines: Fable 5 Extension and Claude Science

Anthropic has extended Fable 5 access through July 19 and launched Claude Science, an AI workbench for scientists. The company continues to attract high-profile talent including Tom Blomfield, Mike Krieger, Andrej Karpathy, and Peter Bailis joining the company.

Grok 4.5 and GPT-5.6 Beat Anthropic in PR Security Review

A benchmark study shows Grok 4.5 and GPT-5.6 outperforming Anthropic models for finding security vulnerabilities in pull requests, highlighting ongoing model competition in specialized AI tasks.

Agent Benchmarking Tool

A new tool has been released to benchmark engineering teams' AI agent maturity in 5 minutes, receiving positive reception on Hacker News.

Ollama Reaches 9M Users

Ollama, the popular open-source local LLM platform, has reached 9 million users following a $65M funding round, cementing its position as the leading local LLM platform for developers.