MCP 服务器

Agentic Software Factory MCP

Turns Cursor into a full software factory with 10 specialist agents (PM, Architect, Backend, Frontend, etc.) as a local MCP server, keeping code on your machine.

README

Agentic Software Factory MCP

Turn Cursor into a full software factory.

10 specialist agents (PM, Architect, Backend, Frontend, Database, AI, Security, QA, DevOps, Code Reviewer) as a local MCP server.

Your code stays on your machine.

p.s. If this saves you time, star the repo - it helps others find it.

Demo: software_architect → database_architect → code_reviewer

<p align="center"><sub>Architect → Database → Code review in one local MCP factory</sub></p>

Why this exists

AI coding tools are powerful, but they often answer as a generic assistant.

This MCP gives you a role-based factory inside Cursor (or Claude Desktop): PM, Architect, Backend, Frontend, Database, AI, Security, QA, DevOps, and Code Reviewer - all local, via Docker or Python.

Install in 3 minutes

Docker (recommended)

git clone https://github.com/tresor228/agentic-software-factory-mcp.git
cd agentic-software-factory-mcp
./scripts/install-docker.sh

Windows PowerShell:

.\scripts\install-docker.ps1

Paste the printed JSON into Cursor → Settings → MCP
Restart Cursor
Confirm the server is green: agentic-software-factory → 7 tools, 10 prompts, 4 resources

Works on Linux, macOS, and Windows (Docker Desktop).

<details> <summary>Advanced: local Python / uv</summary>

./scripts/install.sh      # Python 3.11+
./scripts/install-uv.sh   # no system Python

See docs/installation.md.

</details>

How to use the agents

Important: in Cursor, agents are MCP prompts, not @mentions.

@ usually lists files/docs. Your agents appear in the MCP prompts menu or via / (depends on Cursor version).

Example flow:

1. Prompt: software_architect
   Task: Analyze this repo and propose a scalable architecture.

2. Prompt: database_architect
   Task: Design the schema from existing models.

3. Prompt: code_reviewer
   Task: Review the latest implementation for bugs and security.

Full guide: docs/usage.md · Agent list: AGENTS.md

What's inside

Agent	What you get
`product_manager`	Specs, stories, prioritization
`software_architect`	Architecture, boundaries, trade-offs
`backend_engineer`	APIs, services, server design
`frontend_engineer`	UI structure, UX flows
`database_architect`	Schemas & migrations from code
`ai_engineer`	LLM / RAG pipelines
`security_engineer`	Threats & hardening
`qa_engineer`	Test strategy & edge cases
`devops_engineer`	CI/CD & observability
`code_reviewer`	Quality + security review

Tools: filesystem (read/list/search) · GitHub (public without token)
Knowledge: coding, architecture, security, database rules as MCP resources
Config: none required to start

Security (read this)

Default mount is ~/Projects when it exists; otherwise home (with a warning).
Prefer mounting a projects folder, not your full home (.ssh, .aws, …).
Never paste a GitHub PAT in chat. Use GITHUB_TOKEN in the MCP process env only if you need private repos.

Project layout

src/server.py      MCP entrypoint (stdio)
src/agents/        10 role prompts
src/tools/         filesystem + GitHub
src/knowledge/     shared rules (resources)
scripts/           Docker / Python / uv installers
docs/              installation + usage