Vynn — AI Workflows & Backtesting

This skill provides access to 21 MCP tools for building self-improving AI workflows and running quantitative backtests.

When to Use

• User wants to create or run an AI workflow (multi-step LLM pipelines)
• User wants to backtest a trading strategy using natural language or structured rules
• User wants to optimize prompts in a workflow step using AI-powered analysis
• User wants to schedule workflows to run on a cron schedule
• User wants to optimize a portfolio using mean-variance optimization
• User asks about workflow analytics (success rates, latency, cost)

Prerequisites

Requires the vynn-mcp MCP server to be configured:

pip install vynn-mcp
export VYNN_API_KEY="your_key_here"

Get a free API key at https://the-vynn.com

Available Tools

Workflows

• list_workflows — List all your workflows
• get_workflow(workflow_id) — Get full details of a workflow
• create_workflow(name, description, steps?) — Create a new workflow
• run_workflow(workflow_id, input_text) — Execute a workflow
• get_runs(workflow_id) — View recent run history
• get_run_summary(run_id) — Detailed run results

Self-Improving

• optimize_prompt(workflow_id, step_id) — Get AI-optimized prompt variants
• apply_prompt_optimization(workflow_id, step_id, prompt) — Apply a new prompt
• get_model_recommendation(workflow_id, step_id) — Get model swap suggestions
• set_schedule(workflow_id, cron, input, timezone?) — Schedule automatic runs
• get_schedule(workflow_id) / delete_schedule(workflow_id) — Manage schedules
• create_trigger(workflow_id) / list_triggers(workflow_id) — Webhook triggers
• get_analytics(workflow_id) — Performance analytics

Backtesting

• backtest(strategy, universe?, ...) — Run a backtest with natural language or JSON strategies
• batch_backtest(base_strategy, parameter_grid, ...) — Parameter sweep
• optimize_portfolio(universe, method?, ...) — Portfolio optimization

Utilities

• list_templates / clone_template — Workflow templates
• list_available_tools — Available step tools

Examples

Create and run a workflow:

User: "Create a workflow that researches a company and writes an investment memo"
→ Use create_workflow with steps for research and writing
→ Use run_workflow with the company name as input

Backtest a strategy:

User: "Backtest buying NVDA when RSI drops below 30"
→ Use backtest with strategy="Buy when RSI(14) < 30, sell when RSI(14) > 70" and universe=["NVDA"]

Optimize a workflow:

User: "My workflow results aren't great, can you improve it?"
→ Use get_analytics to check performance
→ Use optimize_prompt on underperforming steps
→ Use get_model_recommendation to check if a different model would help

Guidelines

• Always show backtest results clearly: total return, Sharpe ratio, max drawdown, win rate
• When creating workflows, ask the user what each step should do before creating
• For backtesting, support both natural language ("buy when RSI < 30") and structured JSON formats
• When optimizing, show the user the suggested changes before applying them