Read-only SQL across your PostgreSQL, MySQL, S3-compatible buckets and folders of Parquet, CSV and JSON files — and one query can join across all of them, federated in-process by DuckDB.

npx @dbconvert/stream-mcp "shop=postgres://user:password@host:5432/shop"

Several sources at once, each optionally named:

npx @dbconvert/stream-mcp \
  shop=postgres://user:password@host:5432/shop \
  orders=mysql://user:password@host:3306/orders \
  lake=s3://analytics/exports?region=eu-central-1 \
  /home/you/data

Ask which customers sit in the PostgreSQL database but are missing from the MySQL one, or whether the CSV someone sent last week still matches the live table — it answers with a single query. No copying, no staging table, no pipeline.

In a client's configuration

{
  "mcpServers": {
    "dbconvert": {
      "command": "npx",
      "args": ["-y", "@dbconvert/stream-mcp",
               "shop=postgres://user:password@host:5432/shop"]
    }
  }
}

It cannot write

There is no tool here that updates a row, drops a table or alters a schema — absent, not disabled. Every tool declares readOnlyHint, so a client can check rather than trust. Connection details are given once, at startup, and never travel through a tool call into the conversation or the logs.

What this package contains

Kilobytes. The server itself is a native binary with DuckDB linked in, so it is downloaded from the GitHub release on first use, checked against a SHA-256 that ships inside this package, and cached under ~/.cache/dbconvert-streams/ (%LOCALAPPDATA% on Windows).

Which platforms

Your machine	`npx @dbconvert/stream-mcp`	Docker image
Linux x64	yes	yes
Windows x64	yes	yes, through Docker Desktop
macOS (Intel or Apple silicon)	not yet	yes, through Docker Desktop
Linux or Windows on ARM	not yet	not yet

A native macOS build is planned. Until it exists, macOS users run the same server as a container — it behaves identically, because it is the same binary built for Linux:

docker run -i --rm slotix/stream-mcp "postgres://user:password@host:5432/db"

On an unsupported platform this launcher says so and points at that command rather than failing with a download error.

S3 credentials come from AWS_ACCESS_KEY_ID / AWS_SECRET_ACCESS_KEY, never from the URL. LOG_FILE defaults to the cache directory — stdout carries the MCP protocol, so the server will not start without somewhere to log.

Documentation: https://streams.dbconvert.com/docs/mcp/standalone

The server is proprietary software, distributed under the DBConvert Streams licence; this launcher is part of the same distribution.