AI Trading Agent Infrastructure

Give AI trading agents a real strategy engine.

AI trading agents need rails: typed components, validation, backtests, deployment paths, and execution logs they can read. Keel exposes all of it through CLI, API, and MCP — so agents can compose strategies, run backtests, deploy live on Hyperliquid, and inspect what happened, all programmatically.

Agents need rails too

What goes wrong when agents have no infrastructure.

Three failure modes — and the structure that fixes each.

Agents without rails

An LLM agent asked to “write a trading strategy” will hallucinate code, invent functions, and skip evaluation. The output looks plausible and runs nowhere.

With Keel

Agents work inside a typed component registry. Every output is a structurally valid pipeline that can be backtested and deployed without rewriting.

Agents without backtests

Agents that propose trades or strategies without running them through realistic backtests produce confidence without evidence.

With Keel

Agents can run multi-asset portfolio backtests with real fees, slippage, and funding through the same engine as the UI. Confidence comes from the simulation.

Agents with custody

Giving an autonomous agent custody of your funds is a recipe for catastrophic loss when it goes off-script.

With Keel

Live deployment requires user authorization. Keel is non-custodial — keys never leave your wallet, and you can revoke access anytime.

Three surfaces, one engine

CLI, REST API, and MCP.

Same component registry, same backtest engine, same deployment path — exposed three ways so agents and humans use the same primitives.

CLI

Python CLI for agents

Component search, strategy authoring, validation, backtest, and deploy from the command line.

# discover components
$ keel components search "momentum"
Found 8 components: ROC, EWMA, RSI, ...
# create from a template
$ keel strategy new my_strat --template momentum
Created my_strat.py (6 components)
$ keel strategy validate my_strat.py
Valid pipeline · 6 steps · output: WeightSeries
REST API

Programmatic access

Strategy CRUD, backtest, deployment, and execution logs via REST. Same surface the UI uses.

# list components
$ GET /api/components?type=signal
24 components · 8 signals, 6 sizers, ...
# run a backtest
$ POST /api/backtests
btr_a8b2c1 · queued
$ GET /api/backtests/btr_a8b2c1
complete · Sharpe 1.47 · MDD -4.3%
MCP

Model Context Protocol

Native tool integration for Claude, Codex, and other MCP clients. Agents call Keel tools directly.

# tools the agent gets
components.search
strategy.create
strategy.validate
backtest.run
backtest.compare
deploy.propose
# live deployment requires user signoff
FAQ

Common questions

What models work with Keel’s MCP server?

Any LLM that supports the Model Context Protocol — Claude, ChatGPT/Codex, and other MCP-compatible clients. The MCP server exposes the same component discovery, validation, and backtest endpoints the CLI uses.

Can agents trade autonomously?

Agents can compose and backtest strategies autonomously. Live deployment requires user authorization — Keel is non-custodial and uses Hyperliquid’s native delegation, so the user signs off on what runs. Agents propose; humans confirm.

Is there an OpenAPI spec?

Yes. The REST API has an OpenAPI spec covering strategy CRUD, backtest, deployment, and execution-log endpoints. The CLI and MCP server are wrappers over the same API surface.

How do I install the MCP server?

The Keel MCP server installs via the SDK and connects to your Keel account using a personal access token. Add it to your MCP-compatible client and the agent gets access to component search, strategy creation, validation, backtest, and deploy tools.

Can my agent deploy live without my approval?

No. Live deployment to Hyperliquid requires explicit user authorization — the agent cannot bypass the non-custodial signing flow. This is a feature, not a limitation.

What can an agent do that a human cannot?

An agent can iterate on strategy variants overnight, run dozens of backtests in parallel, scan the component registry for candidates that match a hypothesis, and surface the top results. The human still decides what to deploy.

What will you build?

Live on Hyperliquid in minutes.

Get started
Non-custodial
Your keys never leave your wallet. Your strategies run on your account — Keel never holds funds.
Same code, backtest to live
The strategy that passed your backtest is the strategy that trades. Same pipeline, no surprises.
Full visibility
See every position, trade, and decision in real time. Pause anytime. Your account, your control.