Summary
Workflows that get rewritten from their own run history.
The unusual pairing is optimisation next to backtesting: a workflow's prompts can be rewritten based on how its runs went, and a strategy can be swept across parameter variations before anything goes live.
What it is
An MCP server for Vynn, a platform for AI workflows that tune themselves. Beyond creating and running workflows it exposes the optimisation loop — prompt rewriting and model recommendations — plus a separate backtesting set for trading strategies.
What you get
- Workflows listed, inspected step by step, created and run
- Run history and per-run summaries
- Prompt optimisation for a single step, and the optimised prompt applied
- Model swap recommendations
- Cron schedules and inbound webhook triggers
- Backtests from natural-language or structured strategies, parameter sweeps, and mean-variance portfolio optimisation
- Workflow templates to clone, the tools attachable to a step, and performance analytics
Requirements
Pip install vynn-mcp and a Vynn API key in VYNN_API_KEY. Keys are issued on signup at the-vynn.com and free-tier keys start with vynn_free_.
Setup effort
One command plus a key — pip install vynn-mcp, then supply credentials