omnibase

Give your AI agent secure access to every database. PostgreSQL, MySQL, SQLite, ClickHouse, SQL Server, and 50+ more through a single MCP server. Works with Claude Code, OpenCode, GitHub Copilot, Cursor, and any MCP-compatible client.

• One server, every database. Connect Postgres, MySQL, ClickHouse, and SQL Server side by side. Your agent picks the right one.
• Secure by default. Read-only permissions, query classification, credential isolation. Agents get guardrails, not footguns.
• Shareable config. One YAML file defines connections, permissions, and custom tools. Commit it, and your whole team gets the same agent capabilities.

1. Add to Claude Code:

claude mcp add omnibase -- npx -y omnibase-mcp@latest

{
  "mcpServers": {
    "omnibase": {
      "command": "npx",
      "args": ["-y", "omnibase-mcp"]
    }
  }
}

2. Create a config file:

npx omnibase-mcp init

Edit omnibase.config.yaml with your database connections (full configuration guide):

connections:
  app:
    dsn: pg://localhost:5432/myapp    # or $DATABASE_URL from env
    permission: read-write
  analytics:
    dsn: clickhouse://localhost:9000/analytics
    permission: read-write
  warehouse:
    dsn: mysql://user:pass@localhost:3306/warehouse
    permission: read-only

That's it. Your agent now has access to 14 database tools, plus any custom tools you define.

git clone https://github.com/itsJeremyMax/omnibase.git
cd omnibase
pnpm install
pnpm build

You: "I just joined this project. Walk me through the data model."

Your agent connects to app, analytics, and warehouse, discovers tables across all three, traces foreign key relationships, and returns a complete summary: table names, column types, row counts, and how everything connects. What used to take hours of reading migration files takes seconds.

Tools used: list_connections, list_tables, get_schema, get_relationships

-- Connect to each database separately, then run:

-- app (Postgres)
SELECT table_name FROM information_schema.tables WHERE table_schema = 'public';
SELECT column_name, data_type, is_nullable FROM information_schema.columns WHERE table_name = '...';
SELECT tc.table_name, kcu.column_name, ccu.table_name AS foreign_table
  FROM information_schema.table_constraints tc
  JOIN information_schema.key_column_usage kcu ON tc.constraint_name = kcu.constraint_name
  JOIN information_schema.constraint_column_usage ccu ON tc.constraint_name = ccu.constraint_name
  WHERE tc.constraint_type = 'FOREIGN KEY';
SELECT schemaname, relname, n_live_tup FROM pg_stat_user_tables;

-- analytics (ClickHouse — different syntax)
SELECT name FROM system.tables WHERE database = currentDatabase();
SELECT name, type, is_in_primary_key FROM system.columns WHERE table = '...';
-- ClickHouse doesn't have foreign keys, so no FK query available

-- warehouse (MySQL — yet another syntax)
SELECT table_name FROM information_schema.tables WHERE table_schema = DATABASE();
SELECT column_name, data_type, is_nullable FROM information_schema.columns WHERE table_name = '...';
SELECT table_name, column_name, referenced_table_name, referenced_column_name
  FROM information_schema.key_column_usage WHERE referenced_table_name IS NOT NULL;

-- Then manually stitch the results together across all three

You: "Compare our product inventory in warehouse with what customers
      actually ordered last quarter in app. What's sitting on shelves?"

Your agent queries warehouse (MySQL) for current stock levels, pulls last quarter's order volumes from app (Postgres), and joins the results by SKU. It identifies 34 products with over 500 units in stock but fewer than 10 orders. You can't write a single query for this because the data lives in two different database engines.

Tools used: list_connections, get_schema, execute_sql (across app and warehouse)

-- Terminal 1: warehouse (MySQL)
SELECT sku, product_name, quantity_on_hand
  FROM inventory
  WHERE quantity_on_hand > 0;
-- Copy results to a spreadsheet or temp file

-- Terminal 2: app (Postgres)
SELECT p.sku, SUM(oi.quantity) AS units_ordered
  FROM order_items oi
  JOIN products p ON oi.product_id = p.id
  JOIN orders o ON oi.order_id = o.id
  WHERE o.ordered_at >= NOW() - INTERVAL '3 months'
  GROUP BY p.sku;
-- Copy results to the same spreadsheet

-- Manually match SKUs across both result sets
-- Filter to high-stock, low-order items
-- Reformat into something useful

You: "The /api/orders endpoint is returning wrong totals. Figure out why."

Your agent reads the route handler, queries the orders table in app, then checks analytics and finds the pre-computed totals are out of sync. It traces the mismatch to a missing category, updates the analytics table, hits the endpoint again, and confirms the response is now correct.

Tools used: get_schema, execute_sql (across app and analytics)

You: "Failed payments spiked 3x in the last hour. What's going on?"

Your agent explores the payments schema on warehouse (read-only), runs analytics on the last hour of transactions, and identifies that all failures share a single payment provider and error code. Not a single row modified. The connection is read-only, so the agent couldn't write even if it tried.

Tools used: get_schema, execute_sql (read-only on warehouse), get_distinct_values

More use cases →

Your agent gets 14 tools out of the box, covering schema exploration, parameterized queries, and execution plan analysis.

Define SQL templates as MCP tools your whole team can use. Add them to your config and commit. Every agent on the project gets the same capabilities, no code required.

tools:
  get_active_users:
    connection: app
    description: "Get all active users"
    sql: "SELECT * FROM users WHERE active = true"

Chain tools together with compose to build multi-step pipelines:

tools:
  active_user_orders:
    connection: app
    description: "Get orders for all active users"
    compose:
      - tool: get_active_users
        as: users
      - sql: "SELECT * FROM orders WHERE user_id IN ({users.id})"
        as: orders

Custom tools live in omnibase.config.yaml alongside your connections. Commit it to your repo and your team shares the same tools, permissions, and database access.

Full custom tools guide →

Any usql-compatible DSN works. Full configuration guide →

Every query is parsed and classified before it reaches your database. Your agent never sees credentials, can't run dangerous functions, and respects the permission level you set.

• Credentials never reach agents. DSNs resolve server-side
• Read-only by default. Explicit opt-in for writes
• Dangerous functions blocked across all engines
• Multi-statement queries rejected
• Table names validated against schema cache

Security deep dive & architecture →

npx omnibase-mcp status           # health dashboard
npx omnibase-mcp tools list       # list custom tools
npx omnibase-mcp upgrade          # upgrade to latest

Full CLI reference →

See CONTRIBUTING.md for setup, testing, and conventions.

pnpm test                          # unit tests
pnpm test:integration              # cross-database tests (needs Docker)
cd sidecar && go test ./... -v     # Go sidecar tests

Apache 2.0. See LICENSE

This software is provided "as is", without warranty of any kind. The authors and contributors are not liable for any damages, data loss, or issues arising from the use of this tool. Omnibase executes SQL queries against real databases. Always review agent-generated queries, use read-only permissions where possible, and test thoroughly before use in production environments. By using this software, you accept full responsibility for its use.