OpenClaw Database Toolkit v2.0

Database operations for AI agents. 5 tools. D1, PostgreSQL, MongoDB. Zero driver installs. Natural language queries.

What's New in v2.0

• DataBridge Protocol -- Unified query interface across D1 (Cloudflare SQLite), PostgreSQL (any host), and MongoDB Atlas. One tool, three databases.
• NLQ Engine -- Natural Language Query: describe what you want in English, get SQL or MQL back plus results. No SQL knowledge required.
• SafeWrite Guard -- Destructive operations (DELETE, DROP, TRUNCATE) require an explicit confirm: true flag. No accidental data loss.
• SchemaCache -- Table schemas are cached for 5 minutes. get_schema is instant after the first call in a session.

Quick Start

{
  "openclaw-database-toolkit": {
    "type": "streamable-http",
    "url": "https://openclaw-database-toolkit-mcp.yagami8095.workers.dev/mcp"
  }
}

Add to Claude Desktop, Cursor, Windsurf, VS Code, or any MCP-compatible client. Zero install. Works immediately.

Detection Triggers

This skill activates when a user says any of these:

• "query my database", "run this SQL", "SELECT from", "D1 query", "database query"
• "insert records", "add rows to table", "bulk insert", "save to database", "write to DB"
• "update records", "UPDATE WHERE", "bulk update", "set this field", "update these rows"
• "show table schema", "what columns exist", "describe this table", "database schema"
• "add a column", "create table", "run migration", "schema migration", "ALTER TABLE"
• "D1 database", "PostgreSQL", "MongoDB Atlas", "Cloudflare D1", "SQL query"

Named Protocols

DataBridge Protocol

Unified connection interface for three database types:

Connection configuration (stored in Pro key):

Query routing:

1. Tool call includes db_alias (e.g., "my-d1", "prod-postgres", "atlas")
2. DataBridge looks up connection config for alias
3. Routes query to appropriate driver (SQLite, pg, MongoDB driver)
4. Normalizes result to unified format: { rows: [...], count: N, duration_ms: N }

Unified result format:

{
  "rows": [{ "id": 1, "name": "task-1", "status": "done" }],
  "count": 1,
  "duration_ms": 42,
  "db_type": "d1",
  "query_used": "SELECT * FROM tasks WHERE status = 'done'"
}

NLQ Engine

Natural Language Query — describe what you want, get results:

Input:  query_database({ db: "my-d1", nlq: "show me all tasks completed in the last 7 days, sorted by most recent" })

NLQ generates: SELECT * FROM tasks WHERE status = 'completed'
               AND completed_at >= datetime('now', '-7 days')
               ORDER BY completed_at DESC

Returns: { rows: [...], generated_sql: "SELECT...", duration_ms: 38 }

NLQ accuracy by complexity:

SafeWrite Guard

Protection against destructive operations:

Tools (5)

query_database -- Query via DataBridge + NLQ Engine

Execute a SQL/MQL query or natural language query against a connected database.

Wrong / Right:

WRONG: Writing raw SQL in an AI agent's context, hoping the database driver is available
RIGHT: query_database({ db: "my-d1", sql: "SELECT * FROM tasks WHERE status = 'queued' LIMIT 10" })
       -> { rows: [{...}, {...}], count: 10, duration_ms: 38, db_type: "d1" }
       -> No driver install. No connection string management. Just the query.

WRONG: Writing SELECT * on a large table without LIMIT
RIGHT: Always use LIMIT in query_database. No limit defaults to LIMIT 100 for safety.
       Use get_task_stats or aggregation queries for counts, not raw row dumps.

insert_records -- Record Insertion with Type Validation

Insert one or many records into a table. DataBridge validates field types against the schema before inserting.

Wrong / Right:

WRONG: Inserting records with wrong types and getting cryptic database errors
RIGHT: insert_records({ db: "my-d1", table: "tasks",
                        records: [{ title: "new task", priority: "high", status: "queued" }] })
       -> Type validation runs against cached schema before the INSERT
       -> Returns: { inserted: 1, ids: ["task-9b2f"] }

WRONG: Inserting 10,000 records in one call (max batch size is 500)
RIGHT: Use batch_size parameter (default 100, max 500) for large inserts.
       insert_records auto-batches if records array exceeds batch_size.

update_records (Pro) -- Record Update via SafeWrite Guard

Update records matching a WHERE clause or natural language condition. SafeWrite requires confirm: true for updates without WHERE.

Parameters: db, table, set (key-value pairs), where (SQL condition or NLQ string), confirm (required for unfiltered updates)

get_schema -- Schema Introspection via SchemaCache

Retrieve full schema for a database or specific table: columns, types, constraints, indexes, and foreign keys.

Returns: tables list (if no table specified) or table detail (column name, type, nullable, default, is_primary_key, is_foreign_key, indexes)

migrate_schema (Pro) -- DDL Migration

Execute CREATE TABLE, ADD COLUMN, CREATE INDEX, or other DDL statements. Returns migration result and updated schema.

Parameters: db, ddl (SQL DDL string), description (human-readable description of what this migration does)

Audit log: All migrations are logged with timestamp, description, and result. Accessible via get_schema({ include_migrations: true }).

Security & Privacy

• Credentials encrypted at rest -- Database connection strings, passwords, and API tokens are encrypted with AES-256 and never returned in API responses.
• Read isolation -- query_database uses read-only credentials when available. Write operations (insert, update, migrate) use separate write credentials you configure.
• No data retention -- Query results are returned in the API response and never stored on our servers. Your data stays in your database.
• Connection aliases scoped to Pro key -- Other users cannot access your database aliases even if they know the alias name.
• SafeWrite is non-bypassable -- Destructive operations cannot be enabled globally. confirm: true must be passed per-call.
• HTTPS only -- All connections are TLS 1.3. HTTP is rejected.

Pricing

Get Pro Key: https://buy.stripe.com/4gw5na5U19SP9TW288

The OpenClaw Intelligence Stack

All 9 servers share one Pro key. $9/mo = 49 tools.