MCP Prompts Server

{ "anthropic_research": [ { "title": "Constitutional AI: Harmlessness from AI Feedback", "authors": "Bai, Y. et al.", "year": "2022", "url": "https://www.anthropic.com/research/constitutional-ai-harmlessness-from-ai-feedback", "key_insight": "Training AI systems through self-improvement using constitutional principles rather than human feedback", "relevance_to_mcp": "Shows how AI agents can be guided by principles, but without wisdom constraints - leading to over-application of patterns" }, { "title": "Agentic Misalignment: How LLMs could be insider threats", "authors": "Anthropic Research Team", "year": "2025", "url": "https://www.anthropic.com/research/agentic-misalignment", "key_insight": "AI agents can engage in sophisticated reasoning to circumvent constraints when facing obstacles to their goals", "relevance_to_mcp": "Demonstrates how AI agents pursue objectives without considering meta-constraints like simplicity" }, { "title": "Specific versus General Principles for Constitutional AI", "authors": "Anthropic Research Team", "year": "2023", "url": "https://www.anthropic.com/research/specific-versus-general-principles-for-constitutional-ai", "key_insight": "General principles like \"do what's best for humanity\" can guide AI behavior, but specific principles provide better fine-grained control", "relevance_to_mcp": "Suggests that AI agents need specific architectural constraints, not just general \"best practices\"" }, { "title": "How we built our multi-agent research system", "authors": "Anthropic Engineering Team", "year": "2025", "url": "https://www.anthropic.com/engineering/built-multi-agent-research-system", "key_insight": "Multi-agent systems burn through 15x more tokens than single agents but excel at parallelizable tasks", "relevance_to_mcp": "Explains why AI agents tend to create multiple repositories - they optimize for parallelization without considering coordination costs" }, { "title": "Collective Constitutional AI: Aligning a Language Model with Public Input", "authors": "Anthropic & Collective Intelligence Project", "year": "2023", "url": "https://www.anthropic.com/research/collective-constitutional-ai-aligning-a-language-model-with-public-input", "key_insight": "Democratic processes can influence AI development, showing differences between expert and public preferences", "relevance_to_mcp": "Highlights the importance of human judgment in AI decision-making, which was missing in MCP Prompts automation" } ], "gpt5_video_insights": [ { "source": "Matthew Berman - How to Make Better Prompts for GPT-5", "url": "https://www.youtube.com/watch?v=EfOjGyctDcQ", "timestamp": "Aug 19, 2025", "key_concepts": [ "Agentic Eagerness - controlling AI decision-making vs direction-taking", "Reasoning Effort parameter - low/medium/high settings", "Tool Preambles - AI explaining its actions during tool calls", "Self-reflection rubrics - AI creating measurement criteria for itself" ] }, { "source": "OpenAI GPT-5 Prompting Guide", "url": "https://cookbook.openai.com/examples/gpt-5/gpt-5_prompting_guide", "referenced_in_video": true, "key_insights": [ "GPT-5 follows instructions with \"surgical precision\" but can be \"more damaging\" if poorly prompted", "Tool call budgets can limit exploration (e.g., \"maximum of 2 tool calls\")", "Responses API provides 4+ point performance gains over chat completions", "Minimal reasoning mode requires more explicit planning in prompts" ] } ], "academic_papers": [ { "title": "RePrompt: Planning by Automatic Prompt Engineering for Large Language Models Agents", "year": "2025", "url": "https://arxiv.org/abs/2406.11132", "key_insight": "AI agents can optimize prompts based on chat history and reflections, without need for final solution checker", "relevance_to_mcp": "Shows how AI agents can iteratively improve their own instructions - potentially leading to over-optimization" }, { "title": "On the Brittle Foundations of ReAct Prompting for Agentic Large Language Models", "year": "2024", "url": "https://arxiv.org/abs/2405.13966", "key_insight": "ReAct-based prompting improvements may be more fragile than claimed, requiring systematic sensitivity analysis", "relevance_to_mcp": "Suggests that AI agent architectural decisions may be based on brittle assumptions about \"best practices\"" }, { "title": "PromptAgent: Strategic Planning with Language Models Enables Expert-level Prompt Optimization", "year": "2023", "url": "https://arxiv.org/abs/2310.16427", "key_insight": "AI agents can engage in trial-and-error exploration to optimize prompts, reflecting on errors and generating feedback", "relevance_to_mcp": "Explains the systematic approach AI agents took in creating MCP Prompts architecture" }, { "title": "LLMs as Method Actors: A Model for Prompt Engineering and Architecture", "year": "2024", "url": "https://arxiv.org/abs/2411.05778", "key_insight": "AI agents should be thought of as \"method actors\" who fully inhabit their assigned roles and contexts", "relevance_to_mcp": "Explains why AI agents given \"architect\" roles systematically applied architectural patterns everywhere" } ] }