概要
What is mcp-server-architect?
mcp-server-architect is an MCP server that acts as an AI Software Architect. It analyzes codebases to generate Product Requirements Documents (PRDs) and provides reasoning assistance for complex coding tasks using an agent-based architecture with GPT‑4o and Gemini 2.5 models.
How to use mcp-server-architect?
Install it with uvx mcp-server-architect (recommended) or pip install mcp-server-architect, then run with required API keys. It exposes two tools: Architect::generate_prd (for PRD generation) and Architect::think (for reasoning assistance). It integrates with Claude Code via MCP.
Key features of mcp-server-architect
- Multi‑model architecture using OpenAI GPT‑4o and Google Gemini
- Intelligent codebase analysis for architectural understanding
- Agent‑based design that autonomously selects tools
- Tools for code reading, web search, and targeted LLM queries
- Comprehensive PRD generation with architectural insights
- Advanced step‑by‑step reasoning for complex coding problems
- Logfire instrumentation for monitoring and debugging
- Simple deployment with
uvx mcp-server-architect
Use cases of mcp-server-architect
- Generate Product Requirements Documents from an existing codebase
- Provide reasoning assistance when an LLM is stuck on a coding task
- Analyze project structure and dependencies for architectural recommendations
- Support developers in planning new features with detailed technical context
FAQ from mcp-server-architect
What API keys are required to run mcp-server-architect?
You need a Google Gemini API key (required) and optionally an OpenAI API key (for GPT‑4o) and an Exa API key (for web search). A Logfire API key is optional for telemetry.
How do I install and run mcp-server-architect?
The quickest method is using uvx: env GEMINI_API_KEY=your_key uvx mcp-server-architect. You can also install with pip install mcp-server-architect and run as a command: env GEMINI_API_KEY=your_key mcp-server-architect.
How can I integrate mcp-server-architect with Claude Code?
Use claude mcp add architect -- env GEMINI_API_KEY=your_key EXA_API_KEY=your_key uvx mcp-server-architect for local scope, or add -s project or -s user for project/user scope. API keys can also be stored in a .env file.
What tools does mcp-server-architect expose?
Two tools: Architect::generate_prd (generates a PRD from a codebase path and task description) and Architect::think (provides reasoning assistance for a stuck LLM given a detailed request).
