DB2TOON MCP

db2toon

A CLI tool that converts database schemas into the Toon schema definition format.

Overview

db2toon connects to a database and extracts schema information (tables, columns, types, constraints, indexes, routines, triggers, and examples), then converts it into the human-readable Toon format for database design documentation and visualization. PostgreSQL, SQLite, DuckDB, MySQL/MariaDB, CockroachDB, Microsoft SQL Server, and Oracle are supported. pg2toon remains a PostgreSQL compatibility command.

Features

• Schema Extraction: Automatically extracts tables, columns, and metadata from PostgreSQL, SQLite, DuckDB, MySQL/MariaDB, CockroachDB, Microsoft SQL Server, and Oracle
• Type Normalization: Simplifies PostgreSQL types (e.g., character varying → varchar)
• Relationship Mapping: Converts foreign key constraints to inline references or multi-column references
• Comment Preservation: Includes comments where the database exposes them; SQLite does not have catalog comments
• Index Documentation: Extracts and documents database indexes
• Database Objects: Preserves supported enums, types, sequences, synonyms, triggers, routines, extensions, materialized views, packages, and other vendor-specific objects in explicit TOON sections
• Cross-Platform: Builds without CGO for Linux, macOS, and Windows (amd64 and arm64)
• DBML Adapter: Converts DBML files (or standard input) into the same TOON format
• Local MCP: Read-only SQL database and plain-text SQL dump schema extraction to compact TOON for MCP clients.

[DB2TOON MCP server]

Installation

From Source

git clone https://github.com/kamil5b/db2toon.gitcd db2toonCGO_ENABLED=0 go build -o output/db2toon ./cmd/db2toonCGO_ENABLED=0 go build -o output/pg2toon ./cmd/pg2toonCGO_ENABLED=0 go build -o output/dbml2toon ./cmd/dbml2toon

From Releases

Download pre-built binaries from the releases page for your platform.

Go package

The module also exposes a public Go API for callers that want the canonical schema model instead of invoking a command:

import (    "context"    "os"
    "github.com/kamil5b/db2toon")
func extract() error {    db, err := db2toon.Extract(context.Background(), db2toon.Request{        Dialect: "postgres",        Dump:    "./schema.sql",        Options: db2toon.Options{ExampleSample: 2},    })    if err != nil {        return err    }    return db2toon.Encode(os.Stdout, db)}

Set exactly one of Request.DB or Request.Dump. Dump contents are parsed offline and never executed. The public API returns *schema.Database, so callers may inspect or transform the model before encoding it.

Usage

LLM tool integration

Build and run the MCP-compatible stdio server:

CGO_ENABLED=0 go build -o output/db2toon-mcp ./cmd/db2toon-mcp./output/db2toon-mcp

The server exposes db2toon.extract_schema. Its required argument is dialect (postgres, sqlite, duckdb, mysql, mariadb, cockroachdb, mssql, sqlserver, or oracle). Provide exactly one of db or dump; optional extraction settings are supplied in an options object. Dump files are parsed offline and never executed. The tool is read-only, uses a 30-second default timeout, and limits responses to 4 MiB. Set options.timeout and options.max_output_bytes to lower limits when needed. Connection strings are never included in tool errors or results.

Basic Usage

./db2toon postgres -db "postgresql://user:password@localhost/dbname"
# SQLite database file./db2toon sqlite -db ./schema.db
# Plain-text SQL dump./db2toon postgres -dump ./schema.sql./db2toon sqlite -dump ./schema.sql./db2toon mysql -dump ./schema.sql./db2toon mssql -dump ./schema.sql./db2toon oracle -dump ./schema.sql
# DuckDB database file (requires libduckdb at runtime)./db2toon duckdb -db ./analytics.duckdb
# Microsoft SQL Server (defaults to dbo)./db2toon mssql -db 'sqlserver://sa:password@localhost:1433?database=app&encrypt=disable'
# Oracle Database; the current schema is used unless -schema/-schemas is set.# go-ora accepts an EZConnect-style URL../db2toon oracle -db 'oracle://app:password@localhost:1521/FREEPDB1'
# Compatibility command; PostgreSQL is selected automatically../pg2toon -db "postgresql://user:password@localhost/dbname"
# Convert DBML, either from a file or standard input../dbml2toon schema.dbmlcat schema.dbml | ./dbml2toon -out schema.toon

Save to File

./db2toon postgres -db "postgresql://user:password@localhost/dbname" -out schema.toon

Include up to two sample rows per PostgreSQL table in the TOON output, using a stable ordering and a reproducible sample seed:

./db2toon postgres -db "postgresql://user:password@localhost/dbname" \  -example-sample=2 -example-sample-ordered=true -seed=42

The default -example-sample=0 omits @example sections.

SQLite and DuckDB also support -example-sample, but currently use a simple LIMIT query. -example-sample-ordered and -seed are currently effective only for PostgreSQL.

Select multiple schemas, include partitioned tables, and change the default 30-second operation timeout with:

./db2toon postgres -db "$DATABASE_URL" -schema audit./db2toon postgres -db "$DATABASE_URL" -schemas public,audit -include-partitioned -timeout 1m
# SQLite and DuckDB default to the `main` schema. SQL Server defaults to `dbo`.# Oracle defaults to the session's CURRENT_SCHEMA../db2toon sqlite -db ./schema.db -schema main./db2toon duckdb -db ./analytics.duckdb -schema analytics./db2toon oracle -db 'oracle://app:password@localhost:1521/FREEPDB1' -schema APP

Flags

• -db string: Database connection URL or local database path; mutually exclusive with -dump
• -dump string: Plain-text SQL dump path; mutually exclusive with -db
• dialect: postgres, sqlite, duckdb, mysql, mariadb, cockroachdb, mssql, sqlserver, or oracle for db2toon; pg2toon always uses PostgreSQL
• -out string: Output file path (optional, defaults to stdout)
• -schema string: A single schema to extract (defaults to public for PostgreSQL, main for SQLite/DuckDB, dbo for SQL Server, and the session CURRENT_SCHEMA for Oracle)
• -schemas string: Comma-separated schemas to extract; cannot be combined with -schema
• -include-partitioned: Include PostgreSQL partitioned tables
• -include-views: Include supported views
• -exclude-tables string: Comma-separated tables to exclude entirely; accepts table or schema.table
• -exclude-example-tables string: Comma-separated tables to exclude from @example sampling
• -exclude-example-fields string: Comma-separated qualified fields to exclude from examples, such as public.users.password_hash
• -example-sample int: Number of sample rows to include per table (defaults to 0)
• -example-sample-ordered: Select sample rows using deterministic ordering for PostgreSQL (defaults to false)
• -seed int: Seed for reproducible PostgreSQL sample selection (defaults to 0; currently ignored by SQLite/DuckDB)
• -timeout duration: Connection and extraction timeout (defaults to 30s)

Dump mode supports plain-text SQL exports for PostgreSQL, SQLite, DuckDB, MySQL/MariaDB, CockroachDB, SQL Server, and Oracle. Common tables, columns, constraints, indexes, comments, and bounded INSERT examples are parsed without executing the dump. PostgreSQL retains native enums, sequences, views, functions, procedures, and triggers, including dollar-quoted routine bodies. MySQL/MariaDB supports DELIMITER-based routine and trigger declarations. SQL Server and Oracle retain supported views, functions, procedures, triggers, sequences, types, synonyms, and selected vendor objects. Complex vendor-specific PL/SQL/ T-SQL bodies are preserved as available statement text; unsupported declarations are ignored rather than executed.

Output Format

The Toon format provides a clean, human-readable schema definition:

@database app {dialect=postgres}
[users]# User accounts table
  id int {pk}  email varchar {req}  name varchar  created_at timestamptz {req}
@indices  idx_email: ON users USING btree (email)
@example[2]{id,email,name,created_at}:  1,[email protected],Alice,2026-01-10T09:00:00Z  2,[email protected],Bob,2026-01-11T10:30:00Z
[posts]# Blog posts
  id int {pk}  user_id int {req} -> users(id)  title varchar {req}  content text  published_at timestamptz
[comments]# Post comments
  id int {pk}  post_id int {req} -> posts(id)  user_id int {req} -> users(id)  content text {req}  created_at timestamptz {req}

Format Elements

• @database name {dialect=dialect}: Source database metadata. Live connections derive the name from the connection string; dump mode uses the dump filename without its extension.
• [TableName]: Table definition
• # comment: Table or column comments
• name type {tags}: Column definition with optional tags
  • {pk}: Primary key
  • {req}: Required (NOT NULL)
  • Multiple tags: {pk,req}
• -> table(column): Foreign key reference (inline for single columns)
• @indices: Section for database indexes
• @enum: Enumerated type values
• @type: User-defined type metadata
• @sequence: Sequence configuration
• @synonym: Alternate object name
• @routine: Function or procedure metadata and definition where available
• @triggers: Table trigger metadata
• @objects: Vendor-specific schema objects, including Oracle materialized views, packages, partitioned tables, scheduler jobs, and database links
• @example[n]{columns}:: Up to n sampled rows from the table
• // comment: Inline column comment

Oracle coverage

Oracle extraction uses user-visible ALL_* catalog views, so the output is limited to objects the connected account can inspect. It supports tables, views, columns and comments, primary/unique/foreign-key/check constraints, independent indexes, triggers, standalone functions and procedures, user sequences, object types, synonyms, materialized views, packages/package bodies, partitioned-table markers, scheduler jobs, and database links. Oracle-generated identity sequences and system-generated NOT NULL checks are omitted because their information is already represented by the column model.

Package/type source bodies, materialized-view refresh settings, detailed partition definitions, grants/roles, VPD policies, tablespace/storage details, specialized spatial/domain index options, and Oracle SQL dump parsing are not yet represented in the canonical model.

Microsoft SQL Server coverage

Microsoft SQL Server extraction uses sys.* catalog views and defaults to the dbo schema unless -schema or -schemas is supplied. It supports tables and views, columns and MS_Description comments, defaults, identity/computed columns, primary/unique/foreign-key/check constraints, independent indexes, triggers, functions/procedures, alias/table types, sequences, synonyms, and sample rows. View definitions and vendor-specific schema objects are emitted in the TOON object sections where applicable.

SQL Server dump parsing is not supported. The current model also does not yet represent partition functions/schemes, filegroups, temporal or memory-optimized table settings, graph tables, full-text/spatial/XML index internals, permissions, extended properties other than descriptions, Agent jobs, or server-level objects.

Requirements

• Go 1.26.0 or later
• PostgreSQL 9.4+ (for JSON aggregation functions), SQLite, DuckDB, MySQL/MariaDB, CockroachDB, Microsoft SQL Server 2022+, or Oracle Database
• A valid database connection string or local database path
• DuckDB also requires a compatible libduckdb shared library at runtime
• Oracle uses the pure-Go go-ora driver and does not require Oracle Instant Client or CGO

License

MIT