DBConvert Streams (Federated SQL) MCP server

DBConvert Streams

Database IDE + migration + real-time CDC — in one workflow.

Query data, move it, keep it in sync, and let your AI assistant see it all

without switching between tools.

If this looks useful, consider giving it a ⭐

Why this exists

Most setups look like this:

• a DB client for queries
• scripts or tools for migration
• a separate CDC pipeline

It works, but it's fragmented.

DBConvert Streams combines these into one workspace.

What it feels like

Think of it as:

DBeaver / DataGrip

• migration tool
• CDC

but without switching tools every time

Example

Run queries across databases and files:

SELECT *
FROM read_parquet('orders.parquet') o
JOIN postgres.public.customers c
  ON o.customer_id = c.id
LIMIT 10;

Then use the same query as a data source — and stream it anywhere.

What you can do

• explore databases, files, and S3
• run SQL across multiple sources
• move data between systems
• keep it in sync with CDC
• connect Claude, Cursor, or Copilot — the AI reads your live schemas, data, and streams (read-only, via MCP)

All in the same workflow.

Note: This is the public home of DBConvert Streams — example configurations, documentation, issue tracking, and release notes. The product itself is proprietary.

Quick Start

Runs anywhere — your laptop, a VPS, or your own infra. No cloud dependency, no vendor lock-in.

Desktop App

Download for Windows, macOS, or Linux — no account required.

AI client extension

Grab dbconvert-streams-.mcpb from Releases and drop it into Claude → Settings → Extensions. It asks for connection strings and folders, and nothing else has to be installed — no DBConvert Streams, no toolchain, no server to run.

postgres://user:password@host:5432/dbname
mysql://user:password@host:3306/dbname
s3://bucket/folder?region=us-east-1

Several sources go in the one field, separated by spaces — shop=postgres://… orders=mysql://… — and a single question can join across all of them.

Add folders of Parquet, CSV or JSON files with the folder picker. Every source becomes read-only tools in your chat, and one question can span several of them at once.

One file covers Windows and Linux — the bundle carries a binary for each and picks the right one. macOS is planned for a later release.

Not a Claude user? The same server runs as a container, and every MCP client that can launch one can use it:

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

Cursor and VS Code can set that up in one click — Add to Cursor · Add to VS Code — VS Code asks for the connection string as you install it; in Cursor you replace the sample one. Folders, S3 keys and the rest: standalone server.

Shortened. The full README is on GitHub.