postgres-mcp
MCP server for PostgreSQL database management, enabling AI-assisted schema exploration, stored procedure analysis, test data generation, and safe transaction management via Claude Desktop.
README
PostgreSQL Manager + MCP Server
A complete Python application for PostgreSQL database management with two modes:
- Desktop GUI (Tkinter) — for direct human use
- MCP Server — for AI-assisted database analysis via Claude Desktop
What It Does
🔌 Connection Management
- Save multiple named PostgreSQL connection profiles in
config.ini - Test connections before saving
- Fields: host, port, database, username, password, SSL mode
🌲 Object Explorer
- Browse: Schemas → Tables → Columns (with types, nullability, PK markers)
- Browse: Schemas → Views
- Double-click a table → auto-generates
SELECT * FROM ... LIMIT 100 - Right-click → Select Top 100, Show Columns
📝 SQL Query Editor
- Syntax highlighting for SQL keywords
- Press F5 to execute (or click Run)
- Query history (session-based, navigate with ↑/↓ buttons)
- Select specific text to run only that portion
📊 Results Grid
- Scrollable table with column headers
- Row count + elapsed time display
- Export to CSV with one click
🧪 SP Analyzer + Test Data Generator
Upload a .sql file containing a stored procedure or function:
- Analyse — extracts SELECT queries, detects referenced tables, infers WHERE conditions
- Generate Preview — creates Faker-based test data respecting types, FKs, enums, UNIQUE
- Insert Test Data — inserts into an open transaction (NOT committed)
- Run your SP manually in the query editor to verify
- Commit or Rollback — on-demand, manual control
🤖 MCP Server (AI-Powered Mode)
When connected to Claude Desktop, Claude can:
- Fetch SP bodies directly from
pg_proc - Execute SELECT queries found in the SP
- Recursively follow SQL strings stored inside DB column values (the killer feature)
- Understand schema, FKs, enums, CHECK constraints
- Generate and insert test data in FK-dependency order
- Wait for you to verify, then commit or rollback
Prerequisites
Mandatory
| Requirement | Version | Check Command |
|---|---|---|
| Python | 3.10+ | python --version |
| pip | any | pip --version |
| PostgreSQL | 10+ (any server) | psql --version |
Optional (for MCP mode)
| Requirement | Purpose |
|---|---|
| Claude Desktop | AI client that connects to our MCP server |
⚠️ Python must be in your system PATH. Run
python --versionin a terminal to verify.
Installation
Option 1: Run the installer script (Windows)
install.bat
This checks for Python, creates a virtual environment, and installs all dependencies.
Option 2: Manual
pip install -r requirements.txt
Dependencies installed:
psycopg2-binary— PostgreSQL driversqlparse— SQL statement parserFaker— realistic test data generationmcp[cli]— Anthropic's MCP SDK for the AI server
Running the Application
Desktop GUI
run.bat
Or manually:
python main.py
MCP Server (for Claude Desktop)
python mcp_server.py
With auto-connect:
python mcp_server.py --host localhost --port 5432 --db mydb --user postgres --password secret
python mcp_server.py --profile local
python mcp_server.py --dsn "postgresql://user:pass@host:5432/db"
Configuring Claude Desktop
- Copy
claude_desktop_config.jsonto%APPDATA%\Claude\claude_desktop_config.json(or merge into your existing config) - Restart Claude Desktop
- The
postgres-mcptools will appear automatically
Example config:
{
"mcpServers": {
"postgres-mcp": {
"command": "python",
"args": ["C:/Balaji/MyProjects/postgres-mcp/mcp_server.py"],
"cwd": "C:/Balaji/MyProjects/postgres-mcp"
}
}
}
Configuration — config.ini
[app]
theme = clam
font_size = 11
row_limit = 1000
default_test_rows = 10
[profile_local]
name = Local PostgreSQL
host = localhost
port = 5432
database = postgres
username = postgres
password =
ssl_mode = prefer
Add more profiles by duplicating the [profile_xxx] section with a unique name.
MCP Tools Reference
| Tool | Description |
|---|---|
connect_to_postgres |
Connect with host/port/db/user/password |
get_connection_status |
Check if connected |
get_sp_body |
Fetch SP/function source from pg_proc |
list_sps |
List all SPs/functions in a schema |
list_tables |
List base tables |
list_views |
List views |
get_view_definition |
Get view SQL (discover real tables behind views) |
get_table_schema |
Full metadata: columns, PKs, FKs, UNIQUEs, CHECKs |
get_fk_reference_values |
Read valid values from FK parent tables |
get_enum_values |
List valid enum labels |
sample_table_data |
Sample existing rows |
execute_query |
Run any SELECT/read SQL |
execute_in_transaction |
Run INSERT/UPDATE/DELETE inside safe test tx |
generate_test_data |
Faker-based row generation |
begin_test_transaction |
Open a write transaction |
insert_rows |
Insert row dicts into the open transaction |
commit_test_data |
Commit permanently |
rollback_test_data |
Roll back all test inserts |
transaction_status |
Check if a test transaction is open |
Troubleshooting
| Issue | Solution |
|---|---|
python not found |
Add Python to system PATH |
psycopg2 install fails |
Install Visual C++ Build Tools, or use psycopg2-binary (already in requirements) |
| MCP server not showing in Claude | Check %APPDATA%\Claude\claude_desktop_config.json path is correct |
| Connection refused | Verify PostgreSQL is running and accepting connections on the configured host:port |
ModuleNotFoundError: mcp |
Run pip install "mcp[cli]>=1.27,<2" |
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。