# Postgres MCP Pro — Index Tuning + Safe SQL Tools

> Postgres MCP Pro is an MCP server for PostgreSQL that runs safe SQL, explains plans, and recommends indexes so agents can tune databases faster.

## Install

Merge the JSON below into your `.mcp.json`:

## Quick Use

1. Install (choose one):
   ```bash
   docker pull crystaldba/postgres-mcp
   # or
   pipx install postgres-mcp
   ```
2. Configure Claude Desktop (`claude_desktop_config.json`):
   ```json
   {
     "mcpServers": {
       "postgres": {
         "command": "postgres-mcp",
         "args": ["--access-mode=unrestricted"],
         "env": {
           "DATABASE_URI": "postgresql://username:password@localhost:5432/dbname"
         }
       }
     }
   }
   ```
3. Verify: ask Claude to `list_schemas`, then run an `explain_query` for a slow SQL statement.

## Intro

Postgres MCP Pro is an MCP server for PostgreSQL that runs safe SQL, explains plans, and recommends indexes so agents can tune databases faster.

- **Best for:** teams that want an AI-safe Postgres copilot for schema exploration, plan analysis, and index tuning
- **Works with:** PostgreSQL, stdio/SSE MCP clients (Claude Desktop, Cursor), Docker or Python 3.12+
- **Setup time:** 10-25 minutes

## Practical Notes

- Quant: GitHub stars and forks indicate broad usage; treat this as a production-grade starting point for DB copilots.
- Quant: start in read-only, then add write access only after you can replay the same query-analysis prompt 3 times with identical outputs.

## Why it matters

Postgres MCP Pro is useful when you want agents to do real database work (plans, health, indexes) without turning your production DB into an unguarded tool.

- Install paths cover Docker and Python package managers, making it easy to standardize across developer machines.
- The repo documents a concrete Claude Desktop `mcpServers` configuration, which reduces integration guesswork.
- The toolset is explicitly positioned around query plans and index tuning rather than only ad-hoc SQL execution.

## Rollout pattern

- Point it at a staging replica first, and validate schema + EXPLAIN flows end-to-end before touching production.
- Adopt a playbook: (1) schema snapshot, (2) EXPLAIN, (3) index hypothesis, (4) verify with a before/after benchmark.
- Keep a human approval rule for any action that changes indexes or constraints.

## Watchouts

Avoid giving unrestricted write access on day 1. Keep credentials scoped, start with read-only, and review any index recommendations with real workload evidence.

### FAQ

**Q: Does it require Docker?**
A: No. The README shows Docker and Python installs (pipx/uv). Use Docker when you want fewer environment surprises.

**Q: How do I keep it safe?**
A: Prefer read-only and a least-privilege DB user. Add writes only with an explicit approval gate.

**Q: What is a good first task?**
A: Ask the agent to list schemas/tables, then run an EXPLAIN for one slow query and propose one index hypothesis.

## Source & Thanks

> Source: https://github.com/crystaldba/postgres-mcp
> License: MIT
> GitHub stars: 2,730 · forks: 299

---

<!-- ZH -->

## 快速使用

1. 安装（二选一）：
   ```bash
   docker pull crystaldba/postgres-mcp
   # 或
   pipx install postgres-mcp
   ```
2. 配置 Claude Desktop（`claude_desktop_config.json`）：
   ```json
   {
     "mcpServers": {
       "postgres": {
         "command": "postgres-mcp",
         "args": ["--access-mode=unrestricted"],
         "env": {
           "DATABASE_URI": "postgresql://username:password@localhost:5432/dbname"
         }
       }
     }
   }
   ```
3. 验证：先让 Claude 调 `list_schemas`，再对一条慢 SQL 跑 `explain_query` 看计划。

## 简介

Postgres MCP Pro 是面向 PostgreSQL 的 MCP Server，提供安全 SQL 执行、EXPLAIN 分析与索引建议，让 Agent 能以更低风险、更高速度定位并优化数据库问题。

- **适合谁：** 希望用“可控、可审计”的方式让 Agent 帮你查库、看执行计划并做索引调优的团队
- **可搭配：** PostgreSQL、支持 MCP 的客户端（Claude Desktop / Cursor 等）、Docker 或 Python 3.12+
- **准备时间：** 10-25 分钟

## 实战建议

- 量化信息：Stars/Forks 代表社区采用度；适合当作“可上生产”的 DB copilot 起点。
- 量化信息：先只读运行；当你能把同一条诊断提示复现 3 次且输出一致后，再逐步开放写权限。

## 为什么值得收录

当你希望 Agent 真的能做数据库工作（执行计划、健康检查、索引优化），又不想把生产库变成“裸奔工具”时，Postgres MCP Pro 的价值会很明显。

- 同时提供 Docker 与 Python 包管理器安装路径，便于团队统一落地。
- README 给出了 Claude Desktop 的 `mcpServers` 配置示例，集成成本更低。
- 能力重点围绕执行计划与索引优化，而不是只做随意 SQL 执行。

## 落地路径

- 先连 staging/只读副本，跑通 schema 与 EXPLAIN 全链路再考虑生产。
- 固定一个调优模板：1) schema 快照 2) EXPLAIN 3) 索引假设 4) before/after 基准验证。
- 任何会改变索引/约束的操作都建议加人工审批闸门。

## 注意事项

不要一上来就开放不受限写权限：先最小权限 + 只读起步，并用真实 workload 证据去验证任何索引建议。

### FAQ

**一定要用 Docker 吗？**
答：不需要。README 同时提供 Docker 与 pipx/uv 的安装方式；想少踩环境坑就用 Docker。

**如何把风险降到最低？**
答：优先只读 + 最小权限 DB 用户；需要写操作时加明确的审批闸门。

**第一件该做什么？**
答：先列出 schema/table，再对一条慢 SQL 跑 EXPLAIN，并让 Agent 给出 1 个索引假设。

## 来源与感谢

> Source: https://github.com/crystaldba/postgres-mcp
> License: MIT
> GitHub stars: 2,730 · forks: 299


---
Source: https://tokrepo.com/en/workflows/postgres-mcp-pro-index-tuning-safe-sql-tools
Author: MCP Hub