AI Clients (MCP)

An MCP (Model Context Protocol) server lets an AI client — Claude, Cursor, VS Code Copilot, Codex — read your databases directly: list what you have, look at a table and a sample of its rows, run a query you asked for. It never writes. Without it you paste schemas and results into a chat window, and they go stale the moment anything changes.

There are two ways to get one, and they suit different people.

Which one is yours

	Start here
You just want your databases in your AI client. Nothing to install or run: you hand the server connection strings — databases, buckets, folders of data files — and it serves them.	Standalone server
You already run DBConvert Streams. The server reads the workspace you have, so the agent sees the same connections and schemas the app does, plus the state of your streams.	Setup

With a workspace, the agent gets what the Explorer gives you — browse the schema, look at rows, run a SELECT, compare two databases. On top of that it can read your streams: the config, the status of a run, its throughput, its recent errors and logs. Starting a stream stays in the app — six tools report on your migrations and CDC, none of them launches one.

Standalone has no stream tools — without DBConvert Streams installed there are no streams to report on. Everything else is identical. It runs three ways — npx, a Claude extension, or a container — all set out on its own page.

Just want to chat with AI inside DBConvert? Use the built-in AI Chat. It needs no MCP registration, but does use one supported agent CLI installed on your computer. This section is for connecting an external AI client or a remote (HTTP) deployment.

Quickstart with a workspace

If DBConvert Streams is installed, pick by where it runs:

Desktop app → stdio: open Settings → AI Clients, click Connect next to your client, and restart it.
Docker or a server → HTTP: point your client at http(s)://<host>/mcp with your API key — see Connect a remote AI client.

Then ask your client List my DBConvert connections. — if the connections you see in Data Explorer come back, you're done.

Without DBConvert Streams, the standalone server is one command: npx @dbconvert/stream-mcp "postgres://…".

→ Full step-by-step, screenshots, and Docker setup: Setup.

No connections yet? An empty list back means MCP is wired up correctly — there's just nothing to show. Add one in the Data Explorer sidebar (+ New Connection) and ask again; see Connection management.

What you can ask

External MCP clients can use the full toolset for database inspection, analytics, federated queries across sources, query review, stream troubleshooting, and cross-source validation. Built-in AI Chat exposes the tools relevant to its current scope. See AI Chat examples and the full tools reference.

What it cannot do

It inspects and advises, and it never writes:

No data or schema changes — no INSERT / UPDATE / DELETE, no DROP / ALTER / CREATE. Only SELECT and WITH … SELECT pass the safety filter; everything else is rejected at the server before it reaches your database.
No connection or credential changes — sources are read from your workspace or from the connection strings the server was started with, and neither is edited.
No stream control in this version — the AI can't start, stop, pause, resume, or reset streams yet. It surfaces status, errors, and the fix; you run the action.

When the AI suggests a change, you (or DBConvert Streams via the UI) run it. See Safety & privacy for the full boundary.

How it works

The server always runs on your machine or your infrastructure, and only ever answers requests your client makes — it never calls out to DBConvert or to your AI provider on its own.

Where it runs	How your client reaches it
Desktop app	a small process beside DBConvert Streams, started on demand by your client over stdio
Docker or a server	a service in your own stack, over HTTP at `/mcp`
Standalone	a process your client starts — through `npx`, the extension, or a container — over stdio

⚡ No restart needed. Edit a connection in the DBConvert Streams UI and the MCP server picks up the change in a fraction of a second — no reconnecting your AI client.

In this section

Without DBConvert Streams

Standalone server — connection strings in, tools out. npx, a one-click Claude extension, or a container.

With a workspace

Setup — wire the server into any client, over stdio or HTTP.
Clients — which client to pick (Cursor, VS Code Copilot, Claude Desktop, OpenCode, Claude Code, Codex, Windsurf, Antigravity).

Either way

Tools reference — every tool, in plain language.
Safety & privacy — what the AI can and cannot do, what it sees, and how secrets are redacted.
Examples — real end-to-end sessions: plain-English SQL and federated validation.
Troubleshooting — fixes for common setup and runtime issues.