Skippr MCP
Short answer: add https://mcp.skippr.io/mcp to your code agent. It is public. No API key. It tells the agent how to install skipprd, write skippr.yml with skipprd connect, and run discover / sync / query. It does not run skipprd, and it never sees your config or credentials.
You still install skipprd on the machine that owns the pipeline. The MCP is the briefing. skipprd is the engine.
Add it to your agentTool catalog →
Add it in Cursor
Save this as .cursor/mcp.json in the project, or ~/.cursor/mcp.json for every project:
{
"mcpServers": {
"skippr": {
"url": "https://mcp.skippr.io/mcp"
}
}
}Reload Cursor. Every other host: MCP for code agents.
Ask it
Install skipprd, then land local JSON into parquet with skipprd connect.You want skippr_install, then skippr_connect for a File source and File sink, then skippr_run for discover and sync --once. The agent runs those commands on your machine. If it invents a connector name, call skippr_connectors.
Engine how-tos stay on Developer Docs.
What to call
| You want | Tool |
|---|---|
Homebrew, install.sh, or pip install skippr | skippr_install |
How skipprd connect writes skippr.yml | skippr_config |
DATA_DIR, cron, containers, WAL on disk or S3 | skippr_deploy |
| Catalog of sources, destinations, schema sinks | skippr_connectors |
skipprd connect argv for one connector | skippr_connect |
| discover, sync, query, serve, doctor | skippr_run |
| Logs, ALTER, one writer | skippr_operate |
transform.ops, partitions, aggregates | skippr_transforms |
Iceberg ∪ WAL SQL, STREAM, skipprd serve | skippr_query |
migrations/*.sql | skippr_migrate |
WAL, crash recovery, DATA_DIR | skippr_wal |
| Exactly-once, CDC, landing policy, schema evolution | skippr_correctness |
| Failed sync, deadletters, missing connector | skippr_debug |
| Cursor / Claude / VS Code snippets | skippr_agent_setup |
| Canonical skippr.io URL | skippr_docs |
Full arguments: MCP tools. Wire format: MCP protocol.
What it will not do
It will not run skipprd for you. It will not invent a connector that is not in the catalog. Secret fields stay ${NAME}. Plaintext passwords are refused.
If the agent needs a command, it runs it on your machine, against your DATA_DIR.