MCP clients
DBConvert Streams' MCP server works with any MCP-compliant AI client. This page helps you pick one — for the actual connect steps, see Setup.
Every client below sees the same workspace through the same read-only tools: saved connections, schemas, tables, files, and stream state. What differs is the client's interface (IDE, chat, terminal) and which model provider processes what the AI reads — see Safety & privacy for the full data-handling picture.
Quick pick
| If you... | Use |
|---|---|
| Already work in VS Code | VS Code Copilot |
| Want an AI-first IDE | Cursor |
| Want a chat interface, not an editor | Claude Desktop |
| Want a terminal agent and local/self-hosted models | OpenCode |
| Work mostly in the terminal already | Claude Code |
| Already use OpenAI tooling | Codex |
| Already use the Windsurf IDE | Windsurf |
| Already use Google's AI tooling | Antigravity |
VS Code Copilot
Mainstream choice if your team already works in VS Code with GitHub Copilot's agent mode. MCP gives Copilot direct access to your connections and schemas instead of pasted snippets — useful for backend work where database context comes up alongside the code.
Cursor
An AI-first IDE. Useful when database inspection is part of a coding task — checking schema assumptions, reviewing a migration, generating integration code — and you want that without leaving the editor.
Claude Desktop
Not an IDE — a chat interface. Good for DBA-style exploration where you're asking questions rather than editing code: List my DBConvert connections, Describe the orders table, Which tables don't have a primary key?. On Linux, only available via the community build aaddrick/claude-desktop-debian.
Database context the AI reads through MCP is processed by Claude (Anthropic).
OpenCode
An open-source terminal coding agent. The reason to pick it over the others: it can point at a local or self-hosted model (Ollama, vLLM, LM Studio, or any OpenAI-compatible endpoint) instead of a cloud provider — relevant if database metadata leaving your machine is a concern.
OpenCode itself doesn't make data private — that depends on which provider you configure it with. OpenCode + a cloud model sends context to that provider same as any other client; OpenCode + Ollama/vLLM on your own infrastructure can keep it local.
Claude Code
A terminal-first coding agent — useful when you already work from the command line and want an agent that can reason about a repository and make code changes, with database context available via MCP for the same reasoning. As with Claude Desktop, MCP context is processed by Claude.
Codex
OpenAI's coding agent. A fit if your team already uses OpenAI tooling. MCP context the AI reads is sent to OpenAI as part of the chat.
Windsurf
Similar role to Cursor — an AI-first IDE, useful if your team has already standardized on it.
Antigravity
Google's agent-oriented development tool. Works through the same MCP tools as the other clients; reasonable if you're already in the Google AI ecosystem.
Keeping data local
If database metadata leaving your machine is a hard requirement, the only client above that supports it is OpenCode pointed at a local or self-hosted model. Every other client sends whatever the AI reads through MCP to its cloud provider as part of the chat. See Safety & privacy for database-level controls (read-only users, tool restrictions) that apply regardless of client.
Next
- Setup — connect any of these clients, stdio or HTTP.
- Safety & privacy — what the AI sees and where it goes.
- Tools reference — 25+ read-only tools.