Getting started.

This is what you are building toward: your language server, profiled inside VS Code.

Add the crates

cargo add lspkit lspkit-server

Turn on the features you need:

cargo add lspkit --features treesitter
cargo add lspkit-server --features mcp

Implement the engine

Your analysis implements EngineApi. It defines its own report, query and error types, and four operations:

#[async_trait]
pub trait EngineApi: Send + Sync + 'static {
    type Report: Send + Sync + 'static;
    type Query: Send + Sync;
    type Error: StdError + Send + Sync + 'static;

    fn generation(&self) -> Generation;
    async fn report(&self, query: Self::Query, cancel: CancellationToken)
        -> Result<Snapshot<Self::Report>, Self::Error>;
    async fn rescan(&self, scope: RescanScope, progress: Progress)
        -> Result<RescanTicket, Self::Error>;
    fn subscribe(&self) -> GenerationEventStream;
    async fn shutdown(&self) -> Result<(), Self::Error>;
}

• generation increases on every successful rescan.
• report answers a query from a snapshot and stops promptly when cancelled.
• rescan refreshes the analysis and may coalesce concurrent requests.
• subscribe streams generation changes to listeners.
• shutdown drains in-flight work; every later call returns an error.

Serve it

Hand one engine instance to the lspkit-server scaffolding for the editor and, with the mcp feature, to the MCP adapter for agents. Both read the same snapshots.

Learn more

The full behavioural contract is in the spec, and the API reference is on docs.rs.