DataDan

PostgreSQL MCP server that gives Claude Code permission-bound database access via a YAML config.

Point it at your databases, set permissions per schema or table, and Claude can browse structure, query data, run migrations — all within the boundaries you define.

Install

npm install -g datadan

Or use directly with npx (no install needed).

Quick Start

In any project directory with a .env containing PostgreSQL connection strings:

npx datadan init

This will:

1. Scan your .env for postgres:// connection strings
2. Connect and discover all schemas and tables
3. Create datadan.config.yaml with everything locked to read
4. Register DataDan in .mcp.json so Claude Code picks it up automatically

That's it. Restart Claude Code and it has read access to your databases.

Use in Any Project

Option A: Init per project

Run npx datadan init in each project directory. This creates a local config and .mcp.json entry.

Option B: Global config, point from anywhere

Create one config file anywhere:

name: my-databases
default-permission: read
hot-reload: true

databases:
  - name: production
    connection_string: ${DATABASE_URL}
  - name: analytics
    connection_string: ${ANALYTICS_DB_URL}

Then in any project's .mcp.json:

{
  "mcpServers": {
    "datadan": {
      "command": "npx",
      "args": ["datadan", "start", "--config", "/path/to/datadan.config.yaml"]
    }
  }
}

Or set the env var DATADAN_CONFIG to point to your config.

Option C: User-level MCP config

Add to ~/.claude/settings.json to make DataDan available in every project:

{
  "mcpServers": {
    "datadan": {
      "command": "npx",
      "args": ["datadan", "start", "--config", "/home/you/datadan.config.yaml"]
    }
  }
}

Config

datadan.config.yaml controls what Claude can do:

name: my-project
default-permission: read
hot-reload: true

databases:
  - name: main
    connection_string: ${DATABASE_URL}
    # permission: write          # override default for this database

    # schemas:
    #   - name: public
    #     permission: write       # override for this schema
    #     tables:
    #       - name: users
    #         permission: none    # block access to this table

Permissions

Permissions cascade: default-permission -> database -> schema -> table. The most specific override wins.

| Level | What Claude Can Do | |-------|-------------------| | none | No access at all | | read | SELECT, describe tables | | write | read + INSERT, UPDATE | | delete | write + DELETE | | yolo | delete + CREATE, ALTER, DROP, TRUNCATE |

Environment Variables

Connection strings support ${ENV_VAR} syntax. DataDan loads .env from the same directory as the config file. Shell/CI variables take precedence.

Hot Reload

With hot-reload: true, DataDan re-reads the config before each tool call. Change permissions mid-session without restarting.

Tools

Once connected, Claude Code gets these tools:

| Tool | Permission | Description | |------|-----------|-------------| | list_databases | any | List all accessible databases | | list_schemas | any | List schemas in a database | | list_tables | any | List tables in a schema | | describe_table | read | Show columns, types, keys, and constraints | | query | read | Run SELECT statements | | execute | write | Run INSERT / UPDATE statements | | delete | delete | Run DELETE statements | | run_migration | yolo | Run DDL (CREATE, ALTER, DROP, TRUNCATE) | | dbml | read | Export database structure as a DBML file | | register | - | Add a new database connection at runtime |

Dry Run

Preview resolved permissions without starting the server:

npx datadan start --dry-run

 Database | Schema | Table    | Permission
----------+--------+----------+-----------
 main     | public | users    | read
 main     | public | posts    | write
 main     | public | secrets  | none

Schema Sync

DataDan auto-discovers new schemas and tables on startup and periodically during use. If you add a table to your database, it appears in the config automatically — inheriting the nearest permission level.