hfsql-mcp

MCP (Model Context Protocol) server for HFSQL databases via PowerShell ODBC.

Built for Claude Code — works with any MCP-compatible client.

Why?

HFSQL (PC SOFT / WinDev) has quirks that break standard ODBC tooling:

No SELECT without a FROM clause (no DUAL table)
Calculated fields that silently reject INSERT
Date format must be YYYYMMDD
French-accented column/table names
No INFORMATION_SCHEMA support

This server handles all of that transparently via PowerShell + native ODBC.

Prerequisites

Windows with PowerShell 7+ (pwsh)
HFSQL ODBC driver installed (32-bit or 64-bit)
A configured ODBC DSN pointing to your HFSQL server
Node.js 18+

Installation

From npm (recommended)

npm install -g hfsql-mcp

From source

git clone https://github.com/kbdevs/hfsql-mcp.git
cd hfsql-mcp
npm install

Configuration

Claude Code (recommended)

claude mcp add hfsql -- npx hfsql-mcp

Or manually add to your ~/.mcp.json:

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

If installed from source:

{
  "mcpServers": {
    "hfsql": {
      "command": "node",
      "args": ["C:\\path\\to\\hfsql-mcp\\index.js"]
    }
  }
}

Other MCP clients

Any MCP-compatible client can use this server via stdio transport. Point it to the index.js entry point.

ODBC DSN Setup

Before using, configure an ODBC DSN in Windows:

Open ODBC Data Source Administrator (64-bit)
Add a new User DSN with the HFSQL driver
Set: Server Name, Server Port (default 4900), Database
Note the DSN name — you'll use it in hfsql_connect

Tools

| Tool | Description | |------|-------------| | hfsql_connect | Connect to an HFSQL database via ODBC DSN | | hfsql_disconnect | Close a registered connection | | hfsql_list_tables | List all tables with column count and optional row count | | hfsql_describe_table | Get table schema (columns, types, sizes, nullable) | | hfsql_query | Execute SELECT queries, returns JSON (default limit: 100 rows) | | hfsql_execute | Execute INSERT / UPDATE / DELETE, returns affected row count | | hfsql_insert | Smart insert from JSON object — auto-skips calculated fields, auto-formats dates | | hfsql_schema_summary | Full schema overview: all tables with columns, types and row counts |

Usage Examples

Connect

hfsql_connect(id: "mydb", dsn: "kbpro", uid: "admin", pwd: "admin")

List tables

hfsql_list_tables(id: "mydb", with_row_count: true)

Describe a table

hfsql_describe_table(id: "mydb", table: "Articles")

Query

hfsql_query(id: "mydb", sql: "SELECT Reference, Libellé, Prix_Vente_HT FROM [Articles]")

Smart insert

hfsql_insert(id: "mydb", table: "Articles", data: {
  "Reference": "M004",
  "Libellé": "BLÉ TENDRE",
  "Prix_Achat_HT": 4500
})

Calculated fields are detected and skipped automatically. Dates in YYYY-MM-DD format are converted to YYYYMMDD for HFSQL.

Execute UPDATE / DELETE

hfsql_execute(id: "mydb", sql: "UPDATE [Articles] SET Libellé = 'MAÏS EN VRAC' WHERE IDArticle = 1")

Full schema summary

hfsql_schema_summary(id: "mydb")

How It Works

Claude Code  ──MCP/stdio──▶  hfsql-mcp (Node.js)
                                  │
                        PowerShell ODBC via temp file
                                  │
                          HFSQL Server (port 4900)

Each tool call builds a PowerShell script that opens an ODBC connection
The script executes the operation and writes JSON to a UTF-8 temp file
Node.js reads the temp file — no console encoding issues with French characters
Results are returned to the MCP client as structured content

HFSQL Quirks Handled

| Quirk | How it's handled | |-------|-----------------| | No DUAL table | Always uses real table in FROM | | Calculated fields | hfsql_insert auto-detects via IsReadOnly schema flag | | Date format | Auto-converts YYYY-MM-DD → YYYYMMDD | | French accents | UTF-8 temp file output, square-bracket identifiers | | No INFORMATION_SCHEMA | Uses ODBC GetSchema("Tables") + GetSchemaTable() |

License

MIT

MCP Servers