OmniSQL MCP
Enables AI assistants to interact with multiple SQL databases (PostgreSQL, MySQL, SQL Server, SQLite, Trino) through MCP, reusing existing DBeaver connections for queries, schema management, and transactions, with SSH tunneling and read-only mode support.
README
OmniSQL MCP
Universal database MCP server — give AI assistants read/write access to your databases using connections already saved in your local DB client workspace (DBeaver-compatible).
This is a fork of srthkdev/omnisql-mcp that adds SSH tunnel / jump host support. It is not published to npm — build it from this repo (see Installation).
Database Support
Natively supported (direct driver, fast):
- PostgreSQL (via
pg) - MySQL / MariaDB (via
mysql2) - SQL Server / MSSQL (via
mssql) - SQLite (via
sqlite3CLI) - Trino / Presto (via
trino-client)
Postgres-compatible (routed through pg driver automatically):
- CockroachDB, TimescaleDB, Amazon Redshift, YugabyteDB, AlloyDB, Supabase, Neon, Citus
Other databases: Fall back to an external CLI configured via OMNISQL_CLI_PATH. Results vary by CLI.
Features
- Reuses connections already configured in your local DB client workspace — no duplicate setup
- Automatic SSH tunnel / jump host support: transparently connects through the same SSH tunnel and gateway/jump host profile configured on the connection (including chained jump servers), no separate tunnel setup needed
- Native query execution for PostgreSQL, MySQL/MariaDB, SQLite, SQL Server, Trino/Presto
- Connection pooling with configurable pool size and timeouts (pooling not applicable to SQLite or Trino/Presto, which are connectionless per query)
- Transaction support (BEGIN/COMMIT/ROLLBACK)
- Query execution plan analysis (EXPLAIN)
- Schema comparison between connections with migration script generation
- Read-only mode with enforced SELECT-only on
execute_query - Connection whitelist to restrict which databases are accessible
- Tool filtering to disable specific operations
- Query validation to block dangerous operations (DROP DATABASE, TRUNCATE, DELETE/UPDATE without WHERE)
- Data export to CSV/JSON
- Graceful shutdown with connection pool cleanup
Requirements
- Node.js 18+
- A local DB client (DBeaver-compatible) with at least one configured connection
Installation
This fork isn't published to npm — build it from source:
git clone https://github.com/sangameshBB/omnisql-mcp.git
cd omnisql-mcp
npm install
npm run build
Then link the built server so the omnisql-mcp command points at it:
npm install -g .
Do not run
npm install -g omnisql-mcpon its own. That installs the original upstream package from the npm registry, which does not have SSH tunnel / jump host support. You must clone this repo and build it locally, then runnpm install -g .from inside the cloned folder as shown above.
Configuration
Claude Desktop
Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS):
{
"mcpServers": {
"omnisql": {
"command": "omnisql-mcp"
}
}
}
Claude Code
Add to ~/.claude/settings.json:
{
"mcpServers": {
"omnisql": {
"command": "omnisql-mcp"
}
}
}
Cursor
Add to Cursor Settings > MCP Servers:
{
"mcpServers": {
"omnisql": {
"command": "omnisql-mcp"
}
}
}
Without a global install
If you'd rather not run npm install -g ., point your MCP client directly at the built entry point instead:
{
"mcpServers": {
"omnisql": {
"command": "node",
"args": ["/absolute/path/to/omnisql-mcp/dist/index.js"]
}
}
}
Environment Variables
| Variable | Description | Default |
|---|---|---|
OMNISQL_CLI_PATH |
Path to external DB client CLI (used for unsupported-driver fallback) | Unset |
OMNISQL_WORKSPACE |
Path to local DB client workspace directory | OS default |
OMNISQL_TIMEOUT |
Query timeout (ms) | 30000 |
OMNISQL_DEBUG |
Enable debug logging | false |
OMNISQL_READ_ONLY |
Disable all write operations | false |
OMNISQL_ALLOWED_CONNECTIONS |
Comma-separated whitelist of connection IDs or names | All |
OMNISQL_DISABLED_TOOLS |
Comma-separated tools to disable | None |
OMNISQL_POOL_MIN |
Minimum connections per pool | 2 |
OMNISQL_POOL_MAX |
Maximum connections per pool | 10 |
OMNISQL_POOL_IDLE_TIMEOUT |
Idle connection timeout (ms) | 30000 |
OMNISQL_POOL_ACQUIRE_TIMEOUT |
Connection acquire timeout (ms) | 10000 |
OMNISQL_SSH_PASSWORD |
Fallback SSH password if it can't be read from the workspace | Unset |
OMNISQL_SSH_PASSPHRASE |
Fallback SSH private key passphrase | Unset |
OMNISQL_SSH_PRIVATE_KEY_PATH |
Fallback SSH private key file path | Unset |
Read-Only Mode
Blocks all write operations. The execute_query tool only allows SELECT, EXPLAIN, SHOW, and DESCRIBE statements. Transaction tools are disabled entirely.
{
"mcpServers": {
"omnisql": {
"command": "omnisql-mcp",
"env": {
"OMNISQL_READ_ONLY": "true"
}
}
}
}
Connection Whitelist
Restrict which workspace connections are visible. Accepts connection IDs or display names, comma-separated:
{
"mcpServers": {
"omnisql": {
"command": "omnisql-mcp",
"env": {
"OMNISQL_ALLOWED_CONNECTIONS": "dev-postgres,staging-mysql"
}
}
}
}
Disable Specific Tools
{
"mcpServers": {
"omnisql": {
"command": "omnisql-mcp",
"env": {
"OMNISQL_DISABLED_TOOLS": "drop_table,alter_table,write_query"
}
}
}
}
Available Tools
Connection Management
list_connections- List all database connectionsget_connection_info- Get connection detailstest_connection- Test connectivity
Data Operations
execute_query- Run read-only queries (SELECT, EXPLAIN, SHOW, DESCRIBE only)write_query- Run INSERT/UPDATE/DELETEexport_data- Export to CSV/JSON
Schema Management
list_tables- List tables and viewsget_table_schema- Get table structurecreate_table- Create tablesalter_table- Modify tablesdrop_table- Drop tables (requires confirmation)
Transactions
begin_transaction- Start a new transactionexecute_in_transaction- Execute query within a transactioncommit_transaction- Commit a transactionrollback_transaction- Roll back a transaction
Query Analysis
explain_query- Analyze query execution plancompare_schemas- Compare schemas between two connectionsget_pool_stats- Get connection pool statistics
SSH Tunnel / Jump Host
get_ssh_tunnel_info- Inspect the SSH tunnel / jump host profile associated with a connection (redacted, no secrets)
Other
get_database_stats- Database statisticsappend_insight- Store analysis noteslist_insights- Retrieve stored notes
Security
- Read-only enforcement:
execute_queryonly accepts read-only statements (SELECT, EXPLAIN, SHOW, DESCRIBE, PRAGMA). Write operations must usewrite_query. - Query validation: Blocks DROP DATABASE, DROP SCHEMA, TRUNCATE, DELETE/UPDATE without WHERE, GRANT, REVOKE, and user management statements.
- Connection whitelist: Restrict which connections are exposed via
OMNISQL_ALLOWED_CONNECTIONS. - Tool filtering: Disable any tool via
OMNISQL_DISABLED_TOOLS. - Input sanitization: Connection IDs and SQL identifiers are sanitized to prevent injection.
- Recommendation: For production use, also use a database-level read-only user for defense in depth.
Workspace Format Support
Supports both configuration formats written by DBeaver-compatible DB clients:
- Legacy: XML config in
.metadata/.plugins/org.jkiss.dbeaver.core/ - Modern: JSON config in
General/.dbeaver/
Credentials are automatically decrypted from the workspace credentials-config.json.
SSH Tunnel / Jump Host Support
If a connection has an SSH tunnel (network handler) configured in your DB client — including one or more chained jump servers / gateway hosts — every native query, test_connection, transaction, and pooled connection transparently routes through it. No separate tunnel setup is required: the server opens a local port forward through the same SSH hop chain your DB client would use and connects the native driver (pg, mysql2, mssql) to that local endpoint.
- Supports password, public key, and SSH agent authentication per hop
- Supports chained jump servers (
localhost -> jump host(s) -> final SSH host -> database) - Tunnels are opened once per connection and reused across queries; closed on shutdown
- Use
get_ssh_tunnel_infoto inspect a connection's tunnel/jump host profile (host, port, auth type, jump server count) without exposing any secrets - If a password or key passphrase can't be recovered from the workspace's encrypted credential store, set
OMNISQL_SSH_PASSWORD,OMNISQL_SSH_PASSPHRASE, orOMNISQL_SSH_PRIVATE_KEY_PATHas a fallback
Trino / Presto Support
Trino connections work over HTTPS/HTTP (Basic Auth) using the same host/user/password already saved for the connection. A few Trino-specific notes:
- Catalog/schema are optional. If the connection has no default catalog/schema configured (common when browsing multiple catalogs in DBeaver), queries must fully qualify tables as
catalog.schema.table. list_tablesandget_table_schemaare catalog-agnostic by design (viasystem.jdbc.tables/system.jdbc.columns), so they work without a default catalog — but on a large multi-catalog cluster this scans metadata across every catalog, which can be slow and may return duplicate rows if the same table name exists in more than one catalog/schema. For a fast, unambiguous lookup, useexecute_querywithDESCRIBE catalog.schema.tableinstead.- SSH tunneling (above) works the same way for Trino connections as any other driver.
- Trino has no persistent session/transaction model in this server —
begin_transactionand connection pooling are not available for Trino connections (same as SQLite).
Development
git clone https://github.com/sangameshBB/omnisql-mcp.git
cd omnisql-mcp
npm install
npm run build
npm test
npm run lint
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。