Legacy SQL Architect MCP
MCP server that gives AI agents deep understanding of contemporary or legacy relational databases.
Legacy SQL Architect MCP
MCP server that gives AI agents deep understanding of legacy databases.
Connect Claude Desktop (or any MCP-compatible AI client) to your PostgreSQL, SQL Server, or Oracle database and let the AI inspect schemas, read stored procedures, analyze query performance, trace dependencies, generate documentation, and produce working Java code — all in a single conversation, without you copying a single line of SQL.
Why Legacy SQL Architect MCP?
Works with legacy databases, not just modern ones. Full stored procedure source retrieval, trigger dependency mapping, and impact analysis designed for databases that have been running for years or decades — including large PL/SQL packages that other tools fail on.
Read-only by design. Connects in strict read-only mode. Physically cannot modify your database — no INSERT, no UPDATE, no DELETE. Safe to point at production.
No framework lock-in. No Docker. Pure Java + JDBC. No Spring, no ORM, no infrastructure to maintain. Runs as a lightweight native installer on Windows, macOS, and Linux.
Combines structure + logic + real data in one conversation. The AI inspects the schema, reads the stored procedure that processes orders, then samples real orders to understand why something is broken — without you copy-pasting anything between tools.
The AI understands relationships, not just tables.
dependency_graph and find_impact trace how tables, triggers, procedures, views, and foreign keys connect — so the AI tells you what breaks before you change anything.
Generates working Java code, not pseudocode.
generate_java_dao produces ready-to-compile Entity + Repository classes (plain JDBC, Java 17+) directly from your live schema.
Supported Databases
| Database | Tested version |
|---|---|
| PostgreSQL | 11+ |
| SQL Server | 2016+ (Express, Standard, Enterprise) |
| Oracle | 12c+ |
Available Tools
| Tool | What it does |
|---|---|
inspect_schema | Tables, columns, types, primary keys, foreign keys |
data_sampler | Sample rows with automatic sensitive-column masking |
get_procedure_source | Full source code of stored procedures and functions |
query_plan_expert | Execution plan analysis, full table scan detection |
dependency_graph | FK chains, trigger→procedure call chains, view dependencies |
generate_mermaid_erd | Entity-Relationship Diagram in Mermaid format |
generate_documentation | Complete Markdown documentation for the entire schema |
find_impact | Everything that depends on a given table |
generate_java_dao | Java Entity + Repository classes (plain JDBC) per table |
Quick Start
1. Install
Windows — download and run legacy-sql-architect-mcp-1.0.0.exe
Linux (Debian/Ubuntu)
sudo dpkg -i legacy-sql-architect-mcp-1.0.0.deb
Linux (RHEL/Fedora)
sudo rpm -i legacy-sql-architect-mcp-1.0.0.rpm
macOS / JAR — requires Java 21+
java -jar legacy-sql-architect-mcp.jar
2. Configure the database connection
Edit config.json in the installation directory:
{
"db_type": "postgresql",
"db_url": "jdbc:postgresql://localhost:5432/your_database",
"db_user": "your_user",
"db_password": "your_password",
"db_schema": "public",
"data_sampler_rows": 10,
"data_sampler_mask_sensitive": true
}
db_type accepts: postgresql, sqlserver, oracle
3. Add to Claude Desktop
Edit %APPDATA%\Claude\claude_desktop_config.json (Windows) or ~/Library/Application Support/Claude/claude_desktop_config.json (macOS):
{
"mcpServers": {
"legacy-sql-architect": {
"command": "C:\\Program Files\\LegacySQLArchitectMCP\\LegacySQLArchitectMCP.exe"
}
}
}
Restart Claude Desktop. A hammer icon (🔨) confirms the server is active.
4. Try it
"Inspect the schema and give me an overview of the data model."
"Get the source of all stored procedures and summarize the business logic."
"I need to rename a column in the orders table — use find_impact to tell me everything that would break."
Example Prompts
Inspect the schema, generate a Mermaid ERD, sample the orders table,
read all stored procedures, and produce complete Markdown documentation.
Analyze this query with query_plan_expert:
SELECT * FROM orders WHERE status = 'PENDING' AND created_at > '2024-01-01'
Is it using indexes? What would you recommend?
Generate Java DAO classes for all tables using package com.acme.repository,
then write each file into src/main/java/com/acme/repository/.
Build the dependency graph. When a row is inserted into orders,
what triggers fire and what procedures do they call?
Documentation
- Setup Guide — installation, config.json reference, Claude Desktop setup, troubleshooting
- User Guide — all tools with prompt examples, multi-tool workflows, tips
Building from Source
Requires Java 21 and Maven 3.8+.
# Run tests + build JAR
mvn clean package
# Windows installer (run on Windows, requires Java 21 with jpackage)
JAVA_HOME="c:/Java_21" mvn clean package -P windows-exe -Dmaven.test.skip=true
# Linux DEB (run inside Ubuntu/Debian container or system)
mvn clean package -P linux-deb -Dmaven.test.skip=true
# Linux RPM (run inside RHEL/Fedora container or system)
mvn clean package -P linux-rpm -Dmaven.test.skip=true
Integration tests use embedded PostgreSQL — no Docker, no external database required.
Distribution
Legacy SQL Architect MCP is distributed as native installers — not as a Maven library. There is no dependency to add to your pom.xml. Download the installer for your platform from GitHub Releases and follow the Setup Guide.
License
Apache License 2.0 — see LICENSE
相关服务器
Postgres MCP Server
Provides secure database access to PostgreSQL using the Kysely ORM.
JDBC-MCP
Enables AI assistants to interact with various databases through JDBC connections.
MySQL MCP Server
Provides AI agents with direct access to query, search, and analyze MySQL databases.
Fedspeak MCP Server
Access and analyze Federal Reserve (FOMC) statements.
Yahoo Finance
Access financial data and visualization tools from Yahoo Finance.
Local Context Memory MCP
A production-ready persistent memory system for AI agents, offering searchable memory across sessions with semantic search and support for multiple database backends.
MSSQL MCP Server
Interact with Microsoft SQL Server (MSSQL) databases. List tables, read data, and execute SQL queries with controlled access.
DBeaver MCP Server
Integrates with DBeaver to provide AI assistants access to over 200 database types using existing connections.
MySQL MCP Server
A MySQL database server for AI assistants, enabling full CRUD operations, transaction management, and intelligent rollback.
MCP for Neo4j
Connects to Neo4j graph databases with ability to use GDS functions ( when available), a read only mode , and set the sample size for schema detection