Riley Brown joined the Startup Ideas podcast to provide a practical walkthrough on setting up "agents with tools" (MCPs) within large language models like Claude, with a focus on workflow automation and business productivity.
The discussion covered using Docker, the MCP toolkit, and integrating tools like Notion and Glyph, demonstrating how non-technical users can enable agents to execute sophisticated, contextual tasks.
The session emphasized current friction points, the importance of hands-on experimentation, and highlighted open business opportunities as the ecosystem matures.
Riley previewed the upcoming launch of Vibe Code App and encouraged listeners to explore AI agent workflows firsthand.
Greg: Include Riley’s social links in show notes and promote Vibe Code App.
Riley: Share additional resources and tutorials for setting up MCPs with Docker, Notion, and Glyph.
Listeners: Experiment with MCP setup and workflows, as recommended during the podcast.
Overview: Agents with Tools (MCPs) and Workflow Automation
MCPs ("agents with tools") empower LLMs like Claude to perform tasks by leveraging external tools in an iterative loop, differentiating them from traditional, linear automation.
Using tools such as N8N and Docker’s MCP toolkit, non-technical users can add functionality (e.g., internet search, content generation, Notion access) to AI agents, streamlining business operations.
Adding tools (e.g., Firecrawl, Notion, Glyph) to an agent currently requires setup via Docker and configuration tokens/API keys; this is still a challenging, manual process.
The trend is moving toward making AI tool integration easier and more catalog-based, similar to Zapier’s approach to automation.
Practical Demonstration: Notion and Glyph Integrations
Notion’s integration with MCP allows the AI agent to read, edit, and comment directly within Notion databases, automating content workflows and facilitating knowledge management.
Example: The agent generates and adds short-form content and comments based on hooks and research, directly into Notion databases.
Glyph provides a no-code workflow builder enabling complex, multi-step AI tasks (e.g., generating thumbnails) without external API keys, and supports remixing workflows from other users.
Combining Claude with Docker-powered MCPs and these tools allows for seamless execution of sophisticated, multi-tool prompts from a chat interface.
Workflow Management and Best Practices
Create clear project instructions for AI agents (e.g., through Claude projects or Notion system prompts) to ensure reliable, context-aware task completion.
Carefully manage tool access and account integrations, limiting agent permissions to relevant information and workflows for security and clarity.
Restart the MCP server via Docker whenever integration changes are made to ensure updated tool access for the agent.
Reflections, Limitations, and Future Potential
Existing setup is still “janky” and manually intensive (requiring frequent restarts, key management, and troubleshooting).
Despite friction, early experimentation offers high leverage as these workflows will become much easier and more powerful over time.
Major AI platforms (Anthropic, OpenAI, Google, etc.) are expected to make tool integrations more seamless soon—creating significant opportunities for those who become proficient early.
Vibe Code App Launch
Riley is launching Vibe Code App, which acts as a mobile-first version of "Cursor," enabling users to generate mobile apps simply by describing their idea.
Launch video planned for release tomorrow.
Decisions
No formal decisions were made; the discussion was instructional and exploratory.
Open Questions / Follow-Ups
Will Anthropic or other providers natively integrate external tool support in their main interfaces soon, eliminating the need for workarounds like Docker/MCP?
What startups will emerge to further simplify key management and integration of external tools for AI agents?
Greg and Riley to monitor user feedback and technical improvements as they deploy these agent-based automation workflows.