Reading up on MCP

Paca is an open-source, self-hosted project management platform designed for AI agents and humans to collaborate as equal teammates.

Epiq is a local-first, Git-backed issue tracker with a terminal UI and MCP server for agent integration, requiring no SaaS accounts.

Hyper is an HTTP framework for Bun distributed as source via CLI, generating OpenAPI, typed RPC clients, and MCP servers from single route definitions.

Anthropic's platform-focused strategy—open protocols, terminal tools, and ecosystem—is winning enterprise adoption over OpenAI's benchmark-driven model improvements, despite OpenAI's smarter models.

CLI Printing Press automates generating token-efficient Go CLIs, MCP servers, and Claude Code skills from any API by studying official docs, community tools, and sniffed web traffic.

Apple's enforcement of App Store rules against AI coding apps exposes a fundamental incompatibility between static review and adaptive software, requiring a new distribution paradigm.

Treating AI as a collaborator requires structuring context as infrastructure, encoding taste as config, shifting verification early, delegating bigger tasks, and closing feedback loops to compound improvements.

AI agents will bypass the App Store model because they connect via APIs and open protocols (MCP), shifting value from payment to discovery.

PostHog shares lessons from two years building AI agents, urging teams to validate demand with simpler interfaces like MCP servers before committing to custom agents.

As AI assistants answer queries and complete tasks without opening vendor apps, mobile apps become headless backends, and metrics like traffic and sessions decline long before installs.

OpenAI's Codex agent relies on an orchestration layer handling prompt assembly, context management, and a custom protocol for multi-surface support.

Reflecting on one year at Anthropic, the author observes that breakout products like Claude Code and MCP began as side projects by one or two people.

An agent's usefulness comes from its 'harness'—the code, tools, and orchestration around the model—which must be engineered for tasks like durable storage, code execution, and context management.

CLIs must be redesigned for AI agents as primary users, prioritizing raw JSON payloads, schema introspection, input hardening, and safety rails over traditional human discoverability.

Cursor's support team uses Cursor with MCP servers to collapse code, logs, and team knowledge, achieving 5–10x throughput gains.

The Hacker News discussion weighs the trade-offs between MCP and CLI for AI agent tool calling, finding that each has valid use cases depending on context.

Tool description quality and specificity, especially when not to use a tool, significantly influence whether AI agents select it over alternatives.

Paper is a new design tool built on HTML/CSS with an MCP-connected canvas that syncs design tokens and components between visual work and codebases via agents.

Figma introduces Claude Code integration via MCP, enabling users to send production code into Figma as editable design layers.

Dreamer is a new personal intelligence platform where users create and run AI agents via natural language, treating agents like applications in an operating system with the Sidekick as kernel.

An AI agent plays the open-source Micropolis (SimCity) via a REST API and MCP, with 377 mayors having built 1633 cities.

OpenEnv framework from Meta and Hugging Face reveals tool-using agents fail at multi-step reasoning, ambiguity resolution, and execution quality in realistic calendar environments.

Solo is a native terminal workspace that runs AI coding agents like Claude Code and Codex alongside a developer's full stack, providing unified visibility, auto-restart, and MCP-based agent coordination.

Agent skills (structured prompt instructions) are emerging as a key pattern for making LLMs reliable, but still lack standardization and face real-world system integration hurdles.

Mintlify introduced analytics showing how AI agents interact with documentation, enabling teams to optimize docs for agent consumption using tools like llms.txt and MCP.

The article argues that MCP's value is in steering agents with rich context descriptions, that progressive disclosure is counterproductive, and that subagents may better manage context rot.

An agent limited to only executing code can still accomplish complex tasks by creating and improving its own CLI tools.

Craft Agents is an open-source agent orchestration platform that connects to any API, MCP server, or local file via natural language, with a built-in browser, multimodal model support, and session-based workflows.

MCP servers fail because they are built like REST APIs instead of as user interfaces for AI agents, requiring outcome-oriented design.

Providing AI agents with automated feedback loops (back pressure) enables them to self-correct on long-horizon tasks, increasing engineer leverage.

Andy Pavlo's 2025 database review covers PostgreSQL's continued dominance, emerging trends like vibe coding, and commentary on SQLite, DuckDB, and MCP security risks.

Using AI agents with MCP (Model Context Protocol) enables security teams to synthesize organizational context from multiple data sources for continuous, prioritized threat modeling and detection gap analysis.

Cursor introduces dynamic context discovery, pulling only necessary data into the context window via files, reducing agent tokens by up to 46.9%.

Falling costs of LLM-generated code shift software engineering bottlenecks to infrastructure, CI, human-guided abstractions, and code review in 2026.

A structured, example-driven GitHub guide teaching Claude Code features from slash commands to advanced agents using visual tutorials and copy-paste templates.

Marmot is an open-source data catalog that ships as a single Go binary backed by Postgres, aiming to simplify data discovery without complex dependencies.