prompt2mcp 0.4.0

Build MCP servers from natural language prompts — local-first, privacy-focused

Generate production-ready MCP servers from natural language prompts.
_{8 templates · 11 APIs · TypeScript & Python · Docker · Web Dashboard · Claude Desktop integration}

---

MCP Factory

MCP Factory is a CLI + web tool that takes a plain English description and generates a fully working Model Context Protocol server — with tools, types, error handling, validation, code review, and best practices baked in.

pip install "prompt2mcp[web]"
mcpfactory create "a server that reads CSV files and runs SQL queries on them"

No boilerplate. No manual wiring. Just describe what you want.

---

Table of Contents

• Installation
• Quick Start
• Web Dashboard
• Features
• Claude Desktop Integration
• CLI Reference
• Architecture
• Development
• Contributing
• License

---

Installation

From PyPI (recommended):

# Core CLI only
pip install prompt2mcp

# With web dashboard
pip install "prompt2mcp[web]"

From source:

git clone https://github.com/JatinTaori1904/mcp-factory.git
cd mcp-factory
pip install -e ".[dev,web]"

---

Quick Start

1. Create your first server

mcpfactory create "github repo manager with issues and PRs"

This generates a complete MCP server in ./output/github-repo-manager/ with:

github-repo-manager/
├── src/
│   └── index.ts          # Full MCP server with tools
├── package.json           # Dependencies configured
├── tsconfig.json          # TypeScript config
├── .env.example           # Environment variables needed
├── Dockerfile             # Container-ready deployment
├── .dockerignore          # Docker build exclusions
└── README.md              # Usage instructions

2. Run the generated server

cd output/github-repo-manager
npm install
npm start

2b. Or run with Docker

cd output/github-repo-manager
docker build -t github-repo-manager .
docker run -i --env-file .env github-repo-manager

3. Connect to Claude Desktop

mcpfactory config-add github-repo-manager
# → Automatically adds the server to Claude Desktop's config

That's it — your server is now available as tools in Claude.

---

More Examples

# Python server for Slack
mcpfactory create "slack bot that posts messages and manages channels" --lang python

# Use OpenAI for smarter analysis
mcpfactory create "stripe payment processor with refunds" --provider openai --model gpt-4

# Interactive mode — MCP Factory asks follow-up questions
mcpfactory create "something for files" --interactive

# Custom name and output directory
mcpfactory create "notion wiki manager" --name my-wiki --output ./servers

# Browse all available templates
mcpfactory templates

# See supported APIs with auth details
mcpfactory supported-apis

---

Web Dashboard

MCP Factory includes a built-in web dashboard for managing your servers visually.

mcpfactory web
# → Opens at http://localhost:8000

Dashboard features:

Page	What it does
Home	Overview with stats — servers created, templates, APIs, total tools
Create	Generate servers from the browser with real-time feedback
API Registry	Browse all 11 supported APIs with auth info and free tier status
Config	Manage Claude Desktop integration — add/remove servers, export config

The dashboard uses a modern glassmorphism UI with dark theme, gradient accents, and smooth animations.

# Custom host/port
mcpfactory web --host 0.0.0.0 --port 3000

---

Features

🧠 Smart Prompt Analysis

A 5-tier LLM pipeline ensures your prompt is always understood:

Tier	Provider	When
1	Ollama (local)	Default — private, free, offline
2	OpenAI	--provider openai
3	Claude	--provider claude
4	Keyword fallback	Auto if all LLMs unavailable
5	Default template	Ultimate safety net

📦 8 Built-in Templates

Template	Use Case	Tools Generated
file-reader	Read, search, and watch files	file_read, file_search, file_watch
database-connector	Query PostgreSQL, MySQL, SQLite	db_query, db_list_tables, db_describe
api-wrapper	Wrap any REST API	api_get, api_post, api_list
web-scraper	Scrape and parse web content	scrape_url, extract_links, parse_html
document-processor	Parse PDFs, DOCX, Markdown	doc_parse, doc_convert, doc_extract
auth-server	OAuth2 / JWT authentication	auth_login, auth_verify, auth_refresh
data-pipeline	ETL: extract, transform, load	pipeline_extract, pipeline_transform, pipeline_load
notification-hub	Send via email, Slack, webhooks	notify_send, notify_broadcast, notify_schedule

🔌 11 APIs with Auto-Setup

When your prompt mentions a supported API, MCP Factory auto-generates pre-built, production-ready tools — not generic wrappers.

API	Auth	Env Variable	Free Tier	Pre-built Tools
GitHub	Token	GITHUB_TOKEN	✅	gh_list_repos, gh_create_issue, gh_get_pr
Slack	Bot Token	SLACK_BOT_TOKEN	✅	slack_post, slack_list_channels, slack_reply
OpenAI	API Key	OPENAI_API_KEY	❌	openai_complete, openai_embed
Stripe	Secret Key	STRIPE_SECRET_KEY	✅ Test	stripe_create_charge, stripe_list_customers
Notion	Integration	NOTION_API_KEY	✅	notion_query_db, notion_create_page
Spotify	OAuth2	SPOTIFY_CLIENT_ID	✅	—
Google	API Key	GOOGLE_API_KEY	✅ Ltd	—
Twitter	Bearer	TWITTER_BEARER_TOKEN	✅ Ltd	—
Discord	Bot Token	DISCORD_BOT_TOKEN	✅	discord_send, discord_list_channels
Linear	API Key	LINEAR_API_KEY	✅	—
Jira	API Token	JIRA_API_TOKEN	✅	—

💬 Interactive Mode

For vague prompts, MCP Factory asks targeted follow-up questions:

mcpfactory create "build something for files" --interactive

# MCP Factory asks:
# 1. What file types? (code, documents, data files, all)
# 2. What operations? (read, write, search, all)
# → Generates a refined, specific server

✅ MCP Best Practices

Every generated server follows the official MCP best practices:

• Tool annotations — readOnlyHint, destructiveHint, idempotentHint, openWorldHint
• Prefixed naming — Consistent tool prefixes (gh_, slack_, stripe_)
• Schema descriptions — Zod .describe() / Pydantic Field(description=...) on every parameter
• Actionable errors — Errors include what went wrong + what to do next
• Pagination — List operations support cursor / maxResults
• Secrets via env vars — Never hardcoded, always from .env

🔍 LLM Code Review

After generation, an LLM reviews the code for:

• Security issues (hardcoded secrets, injection risks)
• MCP compliance (missing annotations, naming violations)
• Code quality (error handling, edge cases)
• Performance (unnecessary loops, missing pagination)

🌍 TypeScript & Python

mcpfactory create "..." --lang typescript  # default
mcpfactory create "..." --lang python

Both languages get the same quality: proper types, error handling, and MCP compliance.

🔒 Privacy First

• All code stays on your local filesystem
• Metadata in local SQLite (~/.mcpfactory/servers.db)
• No telemetry, no cloud, no tracking
• Use Ollama for 100% offline, private generation

---

Claude Desktop Integration

MCP Factory can automatically configure your servers for Claude Desktop:

# Add a server to Claude Desktop
mcpfactory config-add my-server

# Remove a server
mcpfactory config-remove my-server

# Show current Claude config
mcpfactory config-show

# Export full configuration
mcpfactory config-export

The config-add command:

1. Locates your Claude Desktop config file
2. Adds the server entry with the correct command and args
3. Your server immediately appears in Claude's tool list

---

CLI Reference

COMMAND                        DESCRIPTION
─────────────────────────────────────────────────────────────
mcpfactory create <prompt>     Generate a new MCP server
  -n, --name                     Server name (auto-generated if omitted)
  -l, --lang                     typescript | python (default: typescript)
  -o, --output                   Output directory (default: ./output)
  -p, --provider                 LLM: ollama | openai | claude (default: ollama)
  -m, --model                    Model name (e.g., llama3, gpt-4)
  --interactive / --no-interactive  Follow-up questions for vague prompts

mcpfactory list-servers        List all generated servers
mcpfactory templates           Show available templates
mcpfactory supported-apis      Browse APIs with auth details
mcpfactory info <name>         Show details for a specific server
mcpfactory delete <name>       Delete a server

mcpfactory config-add <name>   Add server to Claude Desktop
mcpfactory config-remove <name> Remove server from Claude Desktop
mcpfactory config-show         Show Claude Desktop config
mcpfactory config-export       Export full configuration

mcpfactory web                 Launch web dashboard
  --host                         Bind address (default: 127.0.0.1)
  --port                         Port number (default: 8000)

---

Architecture

┌─────────────────────────────────────────────────┐
│                  User Input                      │
│         (CLI prompt or Web Dashboard)            │
└──────────────────────┬──────────────────────────┘
                       │
          ┌────────────▼────────────┐
          │  Interactive Refinement  │  ← Follow-up questions
          └────────────┬────────────┘
                       │
          ┌────────────▼────────────┐
          │   LLM Prompt Analyzer   │  ← Ollama / OpenAI / Claude
          │   (Keyword Fallback)    │  ← Offline safety net
          └────────────┬────────────┘
                       │
        ┌──────────────┼──────────────┐
        ▼              ▼              ▼
  ┌──────────┐  ┌──────────┐  ┌──────────────┐
  │   API    │  │ Template │  │  LLM Tool    │
  │ Detector │  │ Matcher  │  │  Generator   │
  │ (11 APIs)│  │(8 templ) │  │              │
  └────┬─────┘  └────┬─────┘  └──────┬───────┘
       │              │               │
       └──────────────┼───────────────┘
                      ▼
          ┌───────────────────────┐
          │    Code Generator     │
          │  ├─ Pre-built API tools│
          │  ├─ LLM-generated tools│
          │  └─ Template code      │
          └───────────┬───────────┘
                      │
          ┌───────────▼───────────┐
          │   Validator + Linter  │  ← Syntax + structure checks
          └───────────┬───────────┘
                      │
          ┌───────────▼───────────┐
          │   LLM Code Reviewer   │  ← Security, quality, MCP
          └───────────┬───────────┘
                      │
        ┌─────────────┼─────────────┐
        ▼             ▼             ▼
  ┌──────────┐ ┌───────────┐ ┌──────────┐
  │  File    │ │  SQLite   │ │  Claude  │
  │  System  │ │  Storage  │ │  Config  │
  └──────────┘ └───────────┘ └──────────┘

---

Development

# Clone and install
git clone https://github.com/JatinTaori1904/mcp-factory.git
cd mcp-factory
pip install -e ".[dev,web]"

# Run all tests (142 tests)
pytest

# Run with verbose output
pytest -v

# Lint
ruff check .

Project Structure

mcp-factory/
├── mcp_factory/
│   ├── cli/              # Typer CLI commands
│   ├── generator/        # Core engine (prompt analysis, code gen, review)
│   ├── storage/          # SQLite database layer
│   └── config/           # Claude Desktop config management
├── web/
│   ├── app.py            # FastAPI application
│   ├── templates/        # Jinja2 HTML templates
│   └── static/           # CSS with glassmorphism theme
├── tests/
│   └── test_generator.py # 142 tests across 21 test classes
├── pyproject.toml        # Build config (Hatchling)
└── README.md

---

Contributing

Contributions are welcome! Here's how to get started:

1. Fork the repository
2. Create a feature branch: git checkout -b feature/my-feature
3. Install dev dependencies: pip install -e ".[dev,web]"
4. Make your changes and add tests
5. Run the test suite: pytest
6. Submit a pull request

---

License

MIT — Built by Jatin Taori