Ways to use SignalAI
Strategies are written in Rust, and the engine runs in the cloud. You reach it through
three surfaces: the web Studio, the Claude MCP connector, and the local
signalai-quant cargo crate. They share the same account, signals, strategies, and
backtests, so anything you build in one shows up in the others. Use whichever fits how you work.
| Surface | Best for | Sign in | Install |
|---|---|---|---|
| Web Studio | Building strategies and running backtests visually | Nothing | |
| Claude (MCP) | Authoring + running inside a Claude chat | Google (OAuth) | Add connector |
| Local (cargo) | Editing + type-checking Rust in your own editor | API token | cargo add signalai-quant |
Web Studio
The fastest way to start, nothing to install. At signal-ai-mu.vercel.app/strategies click Build strategy: describe an idea in plain English, the hosted agent writes the Rust, you set the run params (universe, date range, cash, resolution) in the side panel, and run the backtest inline. Results, equity curve, trades, and stats, render right there and at your backtests. Sign in with Google.
Claude (MCP connector)
Add SignalAI as a connector and author + run from any Claude conversation: "write a momentum strategy for AAPL and backtest it 2015–2024," "list my strategies," "how did that last run do?" Claude writes the Rust and calls SignalAI's tools directly, so you never leave the chat.
Local (cargo crate)
For writing Rust in your own editor. cargo add signalai-quant pulls in the API contract:
your strategy compiles locally against exactly the types the server uses, so cargo build /
cargo test catch errors before you submit. The engine never ships, backtests still run in
the cloud (submit them from the Studio or Claude). See Setup.
cargo add signalai-quant
cargo build # type-check your strategy locally
What works where
No surface does everything; each is shaped for its job. This is what's available today:
| Capability | Web Studio | Claude (MCP) | Local (cargo) |
|---|---|---|---|
| Sign in | ✓ | ✓ | ✓ (token) |
| Write a strategy (Rust) | ✓ | ✓ | ✓ |
| Type-check locally | — | — | ✓ |
| Run a backtest | ✓ | ✓ | — |
| Equity curve + trades + stats | ✓ | ✓ | — |
| List / view strategies | ✓ | ✓ | — |
| List / view backtests | ✓ | ✓ | — |
| Analyze + improve results | ~ | ✓ | — |
| Build a signal (interactive) | ✓ | ✓ | — |
✓ = full, ~ = partial, — = not available. In short: the Studio both writes strategies and runs backtests visually, Claude does the same conversationally and reasons about the results for you, and the local crate is for editing and type-checking Rust in your own editor (you still submit runs from the Studio or Claude).
Choosing a surface
- Just exploring: start with the Web Studio, there's nothing to install.
- You already work in Claude: add the MCP connector.
- You want your own editor and version control: use the cargo crate locally.
You don't have to choose just one. All surfaces stay in sync on a single account, so a strategy you write in chat is instantly available in the Studio and vice-versa.