The Problem: With 10+ frontier models now available (Claude Opus/Sonnet/Haiku, GPT-5/4o, Gemini 2.5, DeepSeek, etc.), developers spend cognitive load and tokens guessing which one to use. The right model for boilerplate is different from the one you need for debugging. But testing both is expensive.

The Solution:

A lightweight MCP server that uses task-aware heuristics to instantly recommend the optimal model based on:

Task type: refactoring, debugging, architecture, boilerplate, logic-heavy work

Performance metrics: speed vs. capability trade-offs

Cost efficiency: right-sizing model selection

Key Features:

Integrates seamlessly into your editor workflow (Claude Desktop, Cursor, Windsurf)

One-line installation: npx architectgbt-mcp

Task-specific recommendations across 8+ AI models

Open source, MIT licensed

Installation:

json { "architectgbt-mcp": { "command": "npx", "args": ["architectgbt-mcp"] } }

Why It Matters:

Model Context Protocol is becoming the standard for AI tooling (see: Windsurf's architecture)

With Anthropic, OpenAI, Google, and others releasing models monthly, having an advisor automating model selection is a productivity multiplier

Developers can focus on coding, not on hyperparameter tuning for which AI provider/model to use

GitHub: 3rdbrain/architectgbt-mcp

Would appreciate feedback from the HN community on use cases, improvements, or related problems in this space.