Book a demo

Strategy Copilot for AI agents

AI agents buy strategies one call at a time: pay per call over x402, no account, no subscription.

Live demoComplete taskTool callsThree prompts to try

mcp.runbot.io · agent sessionIllustrative run
The agent asks Runbot

Find a BTC strategy with an APR of at least 40% and a drawdown under 15%.

Prompts an agent can send
  1. POST/v1/quick-strategy402 · x402
  2. paysigned USDC payment, call replayedheld until success
  3. buildthe AI Agent writes the rules7 rules
  4. backtest1 year of order booksAPR +38.9%
  5. optimize1,240 backtestsAPR +46.2%
  6. donestrategy and export returnedpayment settled
Strategy returnedBTC 1h trend + RSI filterExport ready
APR+46.2%
Max drawdown12.4%
Sharpe1.83

Complete tasks

One prompt states the goal. Runbot builds, backtests and optimizes, and returns the strategy with an export.

Simple tool calls

Build, backtest, optimize, export: the user's own agent reasons, calls each tool and decides what comes next.

MCP API, paid per call

An MCP server and REST endpoints, settled in stablecoins over x402, only when the job succeeds.

The flow

The Copilot lives at mcp.runbot.io, as an MCP server at /mcp and REST endpoints under /v1. An agent describes a strategy, receives a payment request, pays and polls until the job completes.

POST https://mcp.runbot.io/v1/quick-strategy
{ "description": "BTC perpetual, 1h momentum, trailing stop, moderate risk" }

← 402 Payment Required · x402
# replay with a signed payment header
← 200 { "job_id": "…" }
# poll until completed
← completed · baseline vs optimized · export ready
  • Tools: build, backtest, optimize, export, and a one-shot that chains them, each with a status call to poll.
  • Payment: x402, in USDC on Base or USDT on X Layer.
  • Isolation: each paying wallet gets its own account, and one payment buys one build.
  • Backtest window: 7 to 365 days, 30 by default.

Complete tasks or tool calls

  • Complete tasks. One prompt states the goal, and the one-shot call builds, backtests, optimizes and exports.
  • Simple tool calls. The user's own agent calls build, backtest, optimize and export one by one, reads each result and decides the next step.

Prompts an agent can send:

  • "Find a BTC strategy with an APR of at least 40% and a drawdown under 15%."
  • "Build a Polymarket 5-minute Up/Down strategy that never pays more than 0.60 a share."
  • "Build an ETH mean-reversion strategy on Hyperliquid and optimize it for Sharpe."
  • "Backtest this strategy on the last 365 days, then export the best variant."

When a job fails

On Base, the payment settles only when the job succeeds. On X Layer, it settles when the job is accepted; a job that fails or needs clarification can be picked up again through /v1/recover, which is not charged. Service discovery (GET /v1) is not charged either.