Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

Appearance settings

MrAliHasan/mcp-maker

Open more actions menu

Repository files navigation

MCP Maker - MCP Server Generator for Any Data Source

PyPI version Tests codecov Python 3.10+ License: MIT Downloads

MCPWorld Rating

Connect Claude, Cursor, ChatGPT, Grok & DeepSeek to your data in under 60 seconds.

One command. No code. No MCP knowledge required. Generate an MCP server for any MCP client — or chat with your database right in the terminal using OpenAI, Grok, DeepSeek, or 500+ models via OpenRouter.

pip install mcp-maker
mcp-maker init postgres://user:pass@host/mydb
mcp-maker config --install    # wires it into Claude Desktop

Restart Claude - it can now query your database.

Works with:

Databases Spreadsheets & SaaS Files & APIs
PostgreSQL Airtable CSV / TSV / JSON / JSONL
MySQL Notion Excel
SQLite Google Sheets REST APIs (OpenAPI/Swagger)
MongoDB HubSpot
Redis Supabase

379 tests · MIT licensed · Zero runtime dependency on MCP-Maker

🏆 Rated A – High Quality by MCPWorld

"This MCP Server has been rigorously validated and offers comprehensive functionality and a high-quality user experience."

✔ Availability verified   ✔ Practical tools   ✔ User-friendly installation   ✔ Detailed service docs


Why not just ask Claude or Cursor to write an MCP server?

You can — for one table and four tools. MCP-Maker exists for everything after that:

  • 20+ correct tools per table — pagination, sorting, column selection, date-range filters, full-text search, operator filters, aggregations, distinct values, batch inserts, foreign-key joins, CSV/JSON export. Hand-prompted servers rarely get pagination and SQL quoting right, let alone all of it.
  • Schema-aware, not prompt-aware — it inspects your actual database: primary keys, foreign keys, views, column comments, select-field options. Nothing is hallucinated.
  • Safe by default — read-only unless you pass --ops insert,update,delete; per-table RBAC; identifier escaping everywhere; batch limits.
  • Auto-configures Claude Desktopmcp-maker config --install and you're done.
  • A standalone file you own — the generated server has zero runtime dependency on MCP-Maker. Uninstall it; your server keeps working.
  • 379 tests stand behind the generated code — every connector's output is rendered and verified in CI.

What is MCP? The Model Context Protocol is the open standard for connecting AI to external tools and data. MCP-Maker auto-generates a complete MCP server from your data source — you don't write a single line of code.


Quick Start

Generate an MCP Server (for Claude, Cursor, ChatGPT)

pip install mcp-maker

# Generate from your database
mcp-maker init sqlite:///mydata.db

# Connect to Claude Desktop
mcp-maker config --install

# Restart Claude — your AI can now query your data

Chat directly in your terminal

No Claude needed — talk to your SQLite database right from the terminal:

pip install "mcp-maker[chat]"

# Chat with any SQLite database (no init required)
mcp-maker chat sqlite:///mydata.db

💡 init vs chat: init generates a server file for AI clients (Claude, Cursor) and supports all 13 connectors. chat lets you query directly from the terminal — currently supports SQLite only (PostgreSQL and MySQL coming soon).

╭──────────────────────────────────╮
│ 💬 MCP-Maker Chat                │
╰──────────── v0.2.7 ─────────────╯
📊 Connected: 3 tables (users, orders, products)
🔧 12 tools available (read-only)
🧠 Provider: OpenAI (gpt-4o-mini)

You > How many orders were placed this month?
🔧 count_orders(date_from='2026-03-01')
There were 47 orders placed this month.

You > Who is our top customer?
🔧 list_orders(order_by='total', order_dir='desc', limit=1)
Your top customer is Sarah Chen with $12,450 in orders.

LLM Providers: chat supports OpenAI, Grok (xAI), DeepSeek, and OpenRouter (500+ models including Claude, Gemini, Llama):

# OpenAI (default)
mcp-maker chat sqlite:///data.db --api-key sk-xxx

# Grok — auto-detected from the xai- key prefix
mcp-maker chat sqlite:///data.db --api-key xai-xxx --model grok-3-mini

# DeepSeek — use --provider (DeepSeek keys share OpenAI's sk- prefix)
mcp-maker chat sqlite:///data.db --provider deepseek --api-key sk-xxx

# OpenRouter — auto-detected from the sk-or- key prefix
mcp-maker chat sqlite:///data.db --api-key sk-or-xxx --model anthropic/claude-sonnet-4
mcp-maker chat sqlite:///data.db --api-key sk-or-xxx --model google/gemini-2.5-flash

Environment variables also work: OPENAI_API_KEY, XAI_API_KEY, DEEPSEEK_API_KEY, OPENROUTER_API_KEY (the variable a key comes from selects the provider).


How It Works

Your Data Source          MCP-Maker              Output
┌──────────────┐    ┌──────────────────┐    ┌─────────────────────────┐
│ SQLite       │    │                  │    │ 📄 mcp_server.py        │
│ PostgreSQL   │    │                  │    │   ↳ Editable, yours     │
│ MySQL        │───▶│  mcp-maker init  │───▶│                         │
│ Airtable     │    │                  │    │ ⚙️  _autogen_tools.py    │
│ Google Sheets│    │  (auto-inspect)  │    │   ↳ list_users()        │
│ Notion       │    │  (auto-generate) │    │   ↳ search_orders()     │
│ CSV/JSON     │    │                  │    │   ↳ join_tasks_users()   │
│ +6 more      │    └──────────────────┘    └─────────────────────────┘
└──────────────┘                            Standalone Python file ✅
                                            Works forever, even after
                                            uninstalling MCP-Maker

MCP-Maker generates a standalone Python file. No runtime dependency on MCP-Maker — uninstall it after generation and your server keeps running.


Supported Connectors (13)

Connector URI Format Auth Install
SQLite sqlite:///my.db Built-in
Files (CSV/TSV/JSON/JSONL) ./data/ or ./users.csv Built-in
PostgreSQL postgres://user:pass@host/db DB creds pip install "mcp-maker[postgres]"
MySQL mysql://user:pass@host/db DB creds pip install "mcp-maker[mysql]"
Airtable airtable://appXXXX API key pip install "mcp-maker[airtable]"
Google Sheets gsheet://SPREADSHEET_ID Service acct pip install "mcp-maker[gsheets]"
Notion notion://DATABASE_ID Integration pip install "mcp-maker[notion]"
Excel excel:///path.xlsx pip install "mcp-maker[excel]"
MongoDB mongodb://… or mongodb+srv://… (Atlas) DB creds pip install "mcp-maker[mongodb]"
Supabase supabase://PROJECT_REF API key pip install "mcp-maker[supabase]"
REST API openapi:///spec.yaml API token pip install "mcp-maker[openapi]"
Redis redis://host:6379/0 Password pip install "mcp-maker[redis]"
HubSpot hubspot://pat=TOKEN PAT pip install "mcp-maker[hubspot]"
# Install all connectors at once
pip install "mcp-maker[all]"

Generated Tools

For each table/collection, MCP-Maker generates:

Tool Description
list_{table} Paginated listing with filters, sorting, field selection, date ranges
get_{table} Lookup by primary key
search_{table} Full-text search across string columns
count_{table} Count with optional filters
insert_{table} Insert a single record*(with--ops insert)*
update_{table} Update by ID*(with--ops update)*
delete_{table} Delete by ID*(with--ops delete)*
batch_insert_{table} Bulk insert up to 1,000 records in a transaction
batch_delete_{table} Bulk delete by IDs
aggregate_{table} GROUP BY aggregations (count, sum, avg, min, max)
distinct_{table} Distinct values of any column
filter_{table} Operator-based filtering (eq, gt, lt, contains, …) for file/Excel/Mongo sources
join_{from}_with_{to} Cross-table queries via auto-discovered foreign keys
call_api Generic escape-hatch call for any endpoint (OpenAPI sources)
export_{table}_csv Export to CSV
export_{table}_json Export to JSON

Additional tools based on flags: --semantic (vector search), --webhooks (event hooks), --audit (structured logging).


CLI Reference

# Core
mcp-maker init <source>                       # Generate MCP server
mcp-maker chat <source>                       # Chat with your database (NEW)
mcp-maker serve                               # Run the generated server
mcp-maker inspect <source>                    # Dry run — preview what would be generated

# Configuration
mcp-maker config --install                    # Auto-configure Claude Desktop
mcp-maker env set KEY VALUE                   # Store API keys in .env
mcp-maker env list                            # List stored keys (masked)
mcp-maker list-connectors                     # Show available connectors

# Deployment
mcp-maker deploy --platform railway           # Generate Railway deployment files
mcp-maker deploy --platform render            # Render deployment
mcp-maker deploy --platform fly               # Fly.io deployment

# Generation Options
mcp-maker init <source> --ops read,insert     # Control what the LLM can do
mcp-maker init <source> --tables users,orders # Only expose specific tables
mcp-maker init <source> --async               # Async tools (aiosqlite/asyncpg)
mcp-maker init <source> --auth api-key        # Require MCP_API_KEY for access
mcp-maker init <source> --semantic            # Enable ChromaDB vector search
mcp-maker init <source> --webhooks            # Real-time event notifications
mcp-maker init <source> --audit               # Structured JSON audit logging
mcp-maker init <source> --cache 60            # Cache reads for N seconds
mcp-maker init <source> --no-ssl              # Disable SSL (local dev only)
mcp-maker init <source> --consolidate-threshold 10  # Consolidate large schemas

# Chat Options
mcp-maker chat <source> --api-key sk-xxx      # OpenAI key
mcp-maker chat <source> --api-key sk-or-xxx   # OpenRouter key (auto-detected)
mcp-maker chat <source> --model gpt-4o        # Choose model
mcp-maker chat <source> --provider openrouter # Explicit provider
mcp-maker chat <source> --tables users        # Limit to specific tables

Architecture & Security

Non-Destructive Generation

MCP-Maker generates two files:

  • mcp_server.py — Your editable entry point. Add custom tools, business logic, middleware. Never overwritten on re-generation.
  • _autogen_mcp_server.py — Auto-generated tools. Regenerated safely when you run init again.

Security Features

Feature Description
Credential Isolation Connection strings and API keys loaded from.env — never embedded in generated code
Granular Permissions --ops read (default) prevents writes. Explicitly enable insert, update, delete
API Key Auth --auth api-key gates every tool call behind MCP_API_KEY validation
SSL/TLS by Default PostgreSQL and MySQL connections enforce encrypted transport
SQL Injection Prevention Column whitelist validation on all dynamic queries
Batch Limits Bulk operations capped at 1,000 records to prevent resource exhaustion
Rate Limiting Built-in token bucket throttling for cloud APIs (Airtable, Notion, Sheets)

Schema Versioning

MCP-Maker generates a .mcp-maker.lock file tracking your schema fingerprint. On re-generation, it detects changes (added/removed tables and columns) and displays a color-coded migration diff before updating tools.

Large Schema Handling

For schemas with 20+ tables, the --consolidate-threshold flag switches from per-table tools to consolidated generic tools (e.g., query_database), preventing LLM context window overflow.


MCP Client Compatibility

The generated server works with any MCP-compatible client:

Client Setup
Claude Desktop mcp-maker config --install (automatic)
Cursor Add to Cursor Settings → MCP Servers
Windsurf Add to~/.codeium/windsurf/mcp_config.json
VS Code + Continue Add to Continue's MCP config
ChatGPT Desktop OpenAI MCP support (rolling out)
Any MCP client Runmcp-maker serve and point to it

Installation

# Core (SQLite + Files + CLI)
pip install mcp-maker

# With chat support (OpenAI / Grok / DeepSeek / OpenRouter)
pip install "mcp-maker[chat]"

# With specific connectors
pip install "mcp-maker[postgres]"
pip install "mcp-maker[airtable]"
pip install "mcp-maker[gsheets]"
pip install "mcp-maker[notion]"

# With async support
pip install "mcp-maker[async-sqlite]"
pip install "mcp-maker[async-postgres]"
pip install "mcp-maker[async-mysql]"

# Everything
pip install "mcp-maker[all]"

Requirements: Python 3.10+


📖 Documentation

Guide Description
Getting Started Installation, first server, Claude Desktop setup
CLI & Architecture Reference All commands, env vars, security details

Connector Guides

Each guide includes step-by-step setup, examples, and troubleshooting:

Connector Guide
SQLite docs/sqlite.md
Files (CSV/JSON) docs/files.md
PostgreSQL docs/postgresql.md
MySQL docs/mysql.md
Airtable docs/airtable.md
Google Sheets docs/google-sheets.md
Notion docs/notion.md
Excel docs/excel.md
MongoDB docs/mongodb.md
Supabase docs/supabase.md
REST API (OpenAPI) docs/openapi.md
Redis docs/redis.md
HubSpot docs/hubspot.md
Semantic Search docs/semantic-search.md

Contributing

MCP-Maker is designed for community contributions — each connector is a self-contained PR.

See CONTRIBUTING.md for a step-by-step guide.

git clone https://github.com/MrAliHasan/mcp-maker.git
cd mcp-maker
make install    # Set up dev environment
make check      # Run lint + tests (379 tests)

Security

Found a vulnerability? Please report it privately via SECURITY.md.

License

MIT License · Code of Conduct

About

MCP Maker - MCP server generator. Connect Claude, Cursor, ChatGPT, Grok & DeepSeek to PostgreSQL, MySQL, MongoDB, Airtable, Notion + 13 data sources. Zero code.

Topics

Resources

Code of conduct

Contributing

Security policy

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages

Morty Proxy This is a proxified and sanitized view of the page, visit original site.