How-Tos

Task-oriented guides for doing specific things with vibeD. Each one assumes you already have a running instance (locally via make dev or a production install) and walks through one job end to end.

Connect an AI agent

vibeD exposes its tools over MCP at /mcp (streamable HTTP), so any MCP client can deploy to it. These guides cover the exact client-side config for the most common ones:

Guide	Client	Transport
Connect Claude Desktop	Claude Desktop / claude.ai	Remote connector, or mcp-remote bridge
Connect Claude Code	Claude Code CLI	Native streamable HTTP (claude mcp add)
Connect Codex	Codex CLI, IDE extension, ChatGPT desktop	Native streamable HTTP (codex mcp add)
Connect pi	pi.dev coding agent	pi-mcp-adapter extension

Goose is covered in First deployment.

Before you start: what every client needs

Whichever client you use, you need the same two things:

1. The MCP URL. http://localhost:8080/mcp on a dev install; https://<your-host>/mcp in production.
2. A bearer token, if auth is on. The dev overlay runs with auth disabled, so no token is needed. A production install uses an API key or OIDC; the guides show where each client takes the Authorization: Bearer … header.

Test the endpoint first

If a client reports "failed to connect", rule out vibeD itself before debugging the client:

curl -i http://localhost:8080/readyz          # should be 200
curl -i -X POST http://localhost:8080/mcp \
  -H 'Content-Type: application/json' -H 'Accept: application/json, text/event-stream' \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-03-26","capabilities":{},"clientInfo":{"name":"curl","version":"0"}}}'

A 401 on the second call means auth is enabled and you need to pass -H 'Authorization: Bearer <token>'.

Once connected

Every client ends up calling the same tools. Try:

"Deploy a single-page site that says hello from vibeD."

The agent calls deploy_artifact, vibeD claims a warm sandbox, and the reply contains the app URL. From there, get_artifact_status, get_artifact_logs, update_artifact and delete_artifact cover the rest of the lifecycle; see the tool reference.