AutoDBA MCP 🚀

Zero-database-connection static SQL analysis, automated query optimization, and execution plan diagnostics.
Powered by the Model Context Protocol (MCP) and sqlglot.

AutoDBA MCP allows AI coding assistants (Claude Desktop, Cursor, Antigravity, Windsurf) to analyze SQL queries, eliminate performance anti-patterns, synthesize zero-downtime index migration DDL, and parse execution plans—without requiring database credentials, network access, or exposing proprietary data.

⚡ Key Capabilities

• Zero Database Connection Required: Runs 100% offline via AST parsing (sqlglot). No production DB credentials or connection strings needed.
• Static Anti-Pattern Detection: Identifies non-SARGable column functions, Cartesian joins, cursor loops, SELECT *, leading wildcard LIKE, and unsafe locks (NOLOCK).
• Deterministic AST Rewriter: Automatically refactors anti-patterns into high-performance SQL (e.g., converting YEAR(col) = 2026 into half-open range bounds col >= '2026-01-01' AND col < '2027-01-01').
• Zero-Downtime Index Synthesis: Generates non-blocking DDL (WITH (ONLINE = ON) for SQL Server, CREATE INDEX CONCURRENTLY for PostgreSQL, ALGORITHM=INPLACE, LOCK=NONE for MySQL) based on query predicates and missing index requests.
• Execution Plan Diagnostics: Parses SQL Server .sqlplan ShowPlan XML, PostgreSQL EXPLAIN (FORMAT JSON), and MySQL EXPLAIN FORMAT=JSON to pinpoint table scans, sort spills, and bottleneck operators.
• Dialect Support: Native optimization rules for Microsoft SQL Server (T-SQL), PostgreSQL, and MySQL.

🛠️ MCP Tools

📚 MCP Resources & Prompts

🚀 Quickstart & Configuration

Option 1: Claude Desktop (claude_desktop_config.json)

On macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
On Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "autodba": {
      "command": "uvx",
      "args": ["autodba-mcp"]
    }
  }
}

Option 2: Cursor (~/.cursor/mcp.json or project .cursor/mcp.json)

{
  "mcpServers": {
    "autodba": {
      "command": "uvx",
      "args": ["autodba-mcp"]
    }
  }
}

Option 3: Windsurf (~/.codeium/windsurf/mcp_config.json)

{
  "mcpServers": {
    "autodba": {
      "command": "uvx",
      "args": ["autodba-mcp"]
    }
  }
}

Option 4: Antigravity / Gemini CLI (mcp_config.json)

{
  "mcpServers": {
    "autodba": {
      "command": "uvx",
      "args": ["autodba-mcp"]
    }
  }
}

Option 5: Local Pip / Development

# Run directly with uv
uv run autodba-mcp

# Or install via pip
pip install autodba-mcp
autodba-mcp

💡 Example Usage

1. Analyzing Non-SARGable Predicates

Input:

SELECT l.LeadId, l.Email, l.CreatedDate
FROM dbo.Leads l
WHERE YEAR(l.CreatedDate) = 2026

Diagnostic Output:

# AutoDBA Query Analysis Report (TSQL)

## Overview
- **Complexity Score**: **45/100** (🟡 Moderate)
- **Tables Referenced**: `dbo.Leads`
- **Issues Detected**: 1

## Diagnostic Findings
### [NON_SARGABLE_FUNCTION] Index-Blocking Function on Column — 🔴 CRITICAL
**Finding**: Predicate applies scalar function 'YEAR' to column 'CreatedDate', blocking index seek.
```sql
YEAR(l.CreatedDate) = 2026

💡 Remediation: Rewrite predicate to isolate the column: e.g. col >= '2026-01-01' AND col < '2027-01-01'.

---

### 2. Optimizing Query & Synthesizing Index DDL

**Input:**
```sql
SELECT LeadId FROM dbo.Leads l WHERE YEAR(l.CreatedDate) = 2026

AutoDBA Output:

# AutoDBA Query Optimization Report (TSQL)

## Summary
- **Estimated Workload Improvement**: ~75%
- **Engine**: Deterministic AST Transformation
- **Semantic Verification**: ✅ Passed (All base relations and statement types preserved)
- **Anti-patterns Eliminated**: `NON_SARGABLE_FUNCTION`

## Optimized SQL Query
```sql
SELECT
  l.LeadId
FROM dbo.Leads AS l
WHERE
  l.CreatedDate >= '2026-01-01' AND l.CreatedDate < '2027-01-01'

Recommended Zero-Downtime Index Migration

-- Table: Leads | Rationale: Index supporting equality / range filters: CreatedDate
CREATE NONCLUSTERED INDEX [IX_AutoDBA_Leads_CreatedDate]
ON [dbo].[Leads] ([CreatedDate])
INCLUDE ([LeadId])
WITH (ONLINE = ON, DATA_COMPRESSION = PAGE);

---

## 🛡️ Security & Privacy Guarantees

- **No Remote Calls by Default**: Static analysis and deterministic rewrites execute 100% locally on your machine.
- **Zero Database Access**: AutoDBA never requires DB hostnames, ports, usernames, or passwords.
- **Sanitized Inputs**: Strips non-printable control characters and enforces AST validation before processing.

---

## 🧪 Development & Testing

```bash
# Clone the repository
git clone https://github.com/jeffreyehogg/autodba-mcp.git
cd autodba-mcp

# Create virtual environment and install in editable mode
uv venv --python 3.12
source .venv/bin/activate
uv pip install -e ".[dev]"

# Run test suite
pytest

# Code linting
ruff check .

📄 License

MIT License. See LICENSE for details.