ToolAPI

MRX ToolAPI — connect clients and tools running in the cloud.

ToolAPI is a Rust framework for building client-server applications that communicate over WebSocket connections. It enables clients to invoke remote tools via WebSocket, with support for bidirectional message passing, abort signaling, and strongly-typed parameter passing using a dynamic Value system.

Usage

Add toolapi to your Cargo.toml:

[dependencies]

toolapi = "0.1"

Defining a Tool (Server)

A tool is a function that receives a ValueDict of inputs and a Sender for sending progress messages back to the client:

use toolapi::{ValueDict, Sender, Value, ToolError};

fn my_tool(mut input: ValueDict, mut sender: Sender) -> Result<ValueDict, ToolError> {
    let param: String = input.pop("param")?;

    sender.send("Processing...".to_string())?;

    Ok(ValueDict::from([
        ("output".to_string(), Value::String("done".to_string())),
    ]))
}

#[tokio::main]
async fn main() -> Result<(), std::io::Error> {
    toolapi::run_server(my_tool, None).await
}

The server listens on 0.0.0.0:8080 and accepts WebSocket connections at /tool. An optional HTML string can be served at /.

Calling a Tool (Client)

use toolapi::{call, ValueDict, Value};

let input = ValueDict::from([
    ("param".to_string(), Value::String("hello".to_string())),
]);

let result = call("ws://localhost:8080/tool", input, |msg| {
    println!("{msg}");
    true // return false to abort
});

The callback receives progress messages from the tool. Returning false sends an abort signal.

Core Types

MR-Specific Types

ToolAPI includes domain-specific types for MRI simulation and analysis:

• Signal / Encoding — MR signal data and k-space encoding trajectories
• TissueProperties — T1, T2, T2', ADC parameters
• VoxelPhantom / VoxelGridPhantom / MultiTissuePhantom — phantom representations
• EventSeq / BlockSeq — MRI pulse sequence descriptions

Protocol

Communication uses JSON messages over WebSocket:

• Values(ValueDict) — input/output data
• Message(String) — progress messages from tool to client
• Result(Result<ValueDict, ToolError>) — final result
• Abort — client-requested abort

Disclaimer

This README was generated using Claude Code. No LLMs were used in the writing of the code itself.

License

AGPL — see LICENSE for details.