15:I[981,["658","static/chunks/658-a698693f56d0116a.js","183","static/chunks/183-6beeeb801c274aaa.js","474","static/chunks/app/%5Btype%5D/layout-631c4730b83c8223.js"],"Posts"] 16:T7d5,I built Kortx, an MCP server that lets AI assistants like Claude Code consult with OpenAI's GPT-5 models when they need specialized help. Think of it as giving your AI coding assistant a more experienced colleague to bounce ideas off of.

The problem: When you're coding with Claude Code, sometimes you want strategic feedback on your architecture, alternative approaches to a problem, or help debugging a tricky issue. Claude is great, but having a second AI perspective can be valuable.

Kortx provides four specialized consultation tools:

• think-about-plan: Get strategic feedback on technical plans (clarity, risks, dependencies) • suggest-alternative: Explore different approaches with trade-off analysis • improve-copy: Enhance documentation and messaging for clarity • solve-problem: Debug assistance with root cause analysis

What makes it interesting:

• Uses the GPT-5 Responses API (not Chat Completions) with reasoning effort controls • Automatically gathers relevant context from your codebase via Serena, graph-memory, and CCLSP integrations • Built on the Model Context Protocol, so it works with any MCP-compatible AI assistant • Designed to prevent MCP timeouts with optimized defaults (gpt-5-mini, minimal reasoning)

Technical stack: TypeScript, Zod validation, Pino logging, Docker support, full CI/CD with provenance

It's been really useful in my own development workflow - the architecture review and alternative suggestions have saved me from several poor design choices.

GitHub: https://github.com/effatico/kortx-mcp NPM: npx @effatico/kortx-mcp Docs: https://github.com/effa

Agents 2.0: From Shallow Loops to Deep Agents