# This is Perfect for AI Coding Source: https://www.youtube.com/watch?v=8vvTHnta59A Recap page: https://rapidrecap.app/video/8vvTHnta59A Generated: 2025-12-07T13:04:21.478+00:00 --- ## Quick Overview The video demonstrates how to leverage AI assistants like Claude Code and Cursor, integrated with the Inngest Model Context Protocol (MCP), to automate debugging and testing workflows for serverless functions by directly interacting with the local development environment and monitoring execution traces. **Key Points:** - The Inngest Model Context Protocol (MCP) allows AI assistants to interact with the local Inngest dev server (running at localhost:8288) via a curl command for automated testing and debugging. - The demonstration involves building an AI Research Assistant that uses multiple specialized models (GPT-4 Analyst, Claude Summarizer, Gemini Fact-Checker, Mistral Classifier) in parallel for context enrichment. - The speaker successfully updated the codebase to switch from Vercel AI Gateway to direct OpenAI API usage for all five agents, fixing an issue where agents were using incorrect models (e.g., gpt-4-turbo-preview instead of gpt-4). - The MCP provides eight tools categorized into Event Management (send_event, list_functions, invoke_function), Execution Monitoring (get_run_status, poll_run_status), and Documentation Access (grep_docs, read_doc, list_docs). - Using the MCP tools, the speaker triggered a research query, polled the run status, identified the incorrect model usage through the results, and then used Claude to fix the code by updating model names in `src/lib/ai-models.ts`. - The integration represents a fundamental shift, allowing AI to work alongside the execution environment, providing real-time insight, moving away from treating functions as opaque black boxes. - Future enhancements planned include AI-assisted function optimization based on execution traces, automatic test generation, intelligent error diagnosis, and production monitoring integration. ![Screenshot at 00:11: A terminal session showing puppeteer commands navigating to localhost:3000, taking a screenshot, and then showing the successful navigation status, illustrating the automated UI testing aspect that precedes the MCP demonstration.](https://ss.rapidrecap.app/screens/8vvTHnta59A/00-00-11.png) **Context:** This video serves as a practical demonstration and guide on utilizing Inngest's Model Context Protocol (MCP) integration, which connects AI assistants (specifically Claude Code) to a local Inngest development server. The primary context is overcoming the challenges of testing and debugging asynchronous, event-driven serverless workflows by enabling AI to directly interact with the execution environment and codebase. ## Detailed Analysis The video showcases the power of the Inngest Model Context Protocol (MCP) in streamlining the development workflow for serverless functions, particularly for AI-assisted debugging. Initially, the presenter shows generating three poster concepts using AI coding assistance. The core demonstration then shifts to using the Inngest Dev Server and its MCP tools to debug an AI Agent workflow. The presenter manually sets up the MCP configuration file (`.mcp.json`) to connect Claude Code to the local dev server running at 0.0.0.0:8288. After updating the codebase to exclusively use OpenAI models, a test query is run via `inngeest-dev send_event`. Upon polling the status with `inngeest-dev poll_run_status`, the results reveal that some agents (analyst, summarizer, factChecker, classifier) were incorrectly using older or preview models. The presenter then asks Claude to fix the issue, which involves updating model names in `src/lib/ai-models.ts` and documenting the change in `CLAUDE.md`. The subsequent poll confirms all agents are correctly using 'gpt-4-turbo'. The video emphasizes that MCP provides tools for event management (triggering and listing functions), execution monitoring (checking run status/traces), and documentation access, fundamentally changing debugging from manually probing black boxes to having AI work interactively within the development environment. ### AI-Assisted Code Generation - Creating three different poster concepts in one script - Showing code snippets for drawing crab body paths - Demonstrating the ability to automate creative coding tasks ### Inngest MCP Setup - Demonstrating connecting Claude Code to the local Inngest Dev Server via a JSON configuration in `.mcp.json` - Showing the successful connection status in the CLI - Highlighting configuration scope: user, project, and local settings ### Challenges of Async Testing - Listing issues like functions triggered by external events, running asynchronously, complex state management, and orchestration - Noting that traditional debugging involves manual event triggering, editor/browser switching, and log hunting ### MCP Toolset Overview - Event Management tools include send_event, list_functions, invoke_function - Execution Monitoring includes get_run_status and poll_run_status - Documentation Access includes grep_docs for searching embedded documentation ### AI Agent Debugging Workflow - Using `send_event` to trigger a research query - Using `poll_run_status` to check results - Identifying incorrect model usage (e.g., gpt-4-turbo-preview) reported by agents - Asking Claude to update `src/lib/ai-models.ts` and documentation to enforce OpenAI usage ![Screenshot at 00:01: Code snippet showing the structure for creating a 'Crab body' using path drawing commands to define shape dimensions.](https://ss.rapidrecap.app/screens/8vvTHnta59A/00-00-01.png) ![Screenshot at 00:03: Claude confirming the generation of 3 distinct poster concepts: 'Underwater Realm', 'Nautilus Portal', and a third unnamed concept.](https://ss.rapidrecap.app/screens/8vvTHnta59A/00-00-03.png) ![Screenshot at 00:06: The 'Connectors' screen in the AI interface, displaying various third-party tools that can be integrated with Claude, such as Asana, Canva, Gmail, and Notion.](https://ss.rapidrecap.app/screens/8vvTHnta59A/00-00-06.png) ![Screenshot at 00:11: A terminal output confirming the successful navigation of a Puppeteer instance to localhost:3000, which hosts the Next.js application being tested.](https://ss.rapidrecap.app/screens/8vvTHnta59A/00-00-11.png) ![Screenshot at 02:51: A summary of changes generated by Claude after updating the codebase, showing file modifications in ai-models.ts and documentation updates.](https://ss.rapidrecap.app/screens/8vvTHnta59A/00-02-51.png)