npx skills add ...
npx skills add bobmatnyc/claude-mpm-skills --skill axum
Axum (Rust) web framework patterns for production APIs: routers/extractors, state, middleware, error handling, tracing, graceful shutdown, and testing
npx skills add bobmatnyc/claude-mpm-skills --skill axum
Axum is a Rust web framework built on Hyper and Tower. Use it for type-safe request handling with composable middleware, structured errors, and excellent testability.
✅ Correct: typed handler + JSON response
❌ Wrong: block the async runtime
Handlers are async functions that return something implementing IntoResponse.
✅ Correct: route nesting
Prefer extractors for parsing and validation at the boundary:
Path<T>: typed path paramsQuery<T>: query stringsJson<T>: JSON bodiesState<T>: shared application state✅ Correct: typed path + JSON
If your crate is published as a library in addition to being built as a binary, isolate the HTTP stack to avoid bloating library consumers.
✅ Correct: optional HTTP feature
This way:
cargo add my-lib) get just the core logic without the HTTP overhead.cargo install my-crate works as expected.cargo add my-crate --no-default-features for library use.❌ Wrong: unconditional HTTP dependencies
Use State<Arc<AppState>> and keep state immutable where possible.
✅ Correct: AppState via Arc
IntoResponse)Centralize error mapping to HTTP status codes and JSON.
✅ Correct: AppError converts into response
Use tower-http for production-grade layers: tracing, timeouts, request IDs, CORS.
✅ Correct: trace + timeout + CORS
Terminate on SIGINT/SIGTERM and let in-flight requests drain.
✅ Correct: with_graceful_shutdown
Signal handling above is application-level. In production, your supervisor (systemd, launchd, container orchestrator) controls the actual termination window.
systemd (Linux): Set KillSignal=SIGTERM and TimeoutStopSec=120 (or higher) in your .service file.
launchd (macOS): Use launchctl bootout (sends SIGTERM and waits) instead of launchctl kickstart -k (SIGKILL, truncates in-flight I/O).
Client-side reconnection: Have HTTP clients and MCP bridges connecting to the service implement exponential backoff (starting 200ms, capped at 30s) so brief restarts are transparent and don't cascade errors upstream.
Test routers without sockets using tower::ServiceExt.
✅ Correct: request/response test
std::thread::sleep, blocking I/O inside handlers).unwrap() in request paths; return structured errors instead.async fn handler() {
std::thread::sleep(std::time::Duration::from_secs(1)); // blocks executor
}use axum::{routing::get, Router};
fn router() -> Router {
let api = Router::new()
.route("/users", get(list_users))
.route("/users/:id", get(get_user));
Router::new().nest("/api/v1", api)
}
async fn list_users() -> &'static str { "[]" }
async fn get_user() -> &'static str { "{}" }use axum::{extract::Path, Json};
use serde::{Deserialize, Serialize};
#[derive(Deserialize)]
struct CreateUser {
email: String,
}
#[derive(Serialize)]
struct User {
id: String,
email: String,
}
async fn create_user(Json(body): Json<CreateUser>) -> Json<User> {
Json(User { id: "1".into(), email: body.email })
}
async fn get_user(Path(id): Path<String>) -> Json<User> {
Json(User { id, email: "a@example.com".into() })
}[dependencies]
axum = { version = "0.7", optional = true }
tower-http = { version = "0.5", optional = true }
tokio = { version = "1", features = ["full"] }
[features]
default = ["http-server"]
http-server = ["axum", "tower-http"]
[[bin]]
name = "my-service"
required-features = ["http-server"]# Never do this if the crate is also a library:
axum = "0.7"
tower-http = "0.5"
# Library users now pull in the entire web stackuse axum::{extract::State, routing::get, Router};
use std::sync::Arc;
#[derive(Clone)]
struct AppState {
build_sha: &'static str,
}
async fn version(State(state): State<Arc<AppState>>) -> String {
state.build_sha.to_string()
}
fn app(state: Arc<AppState>) -> Router {
Router::new().route("/version", get(version)).with_state(state)
}use axum::{http::StatusCode, response::IntoResponse, Json};
use serde::Serialize;
#[derive(Debug)]
enum AppError {
NotFound,
BadRequest(&'static str),
Internal,
}
#[derive(Serialize)]
struct ErrorBody {
error: &'static str,
}
impl IntoResponse for AppError {
fn into_response(self) -> axum::response::Response {
let (status, msg) = match self {
AppError::NotFound => (StatusCode::NOT_FOUND, "not_found"),
AppError::BadRequest(_) => (StatusCode::BAD_REQUEST, "bad_request"),
AppError::Internal => (StatusCode::INTERNAL_SERVER_ERROR, "internal"),
};
(status, Json(ErrorBody { error: msg })).into_response()
}
}use axum::{routing::get, Router};
use std::time::Duration;
use tower::ServiceBuilder;
use tower_http::{
cors::{Any, CorsLayer},
timeout::TimeoutLayer,
trace::TraceLayer,
};
fn app() -> Router {
let layers = ServiceBuilder::new()
.layer(TraceLayer::new_for_http())
.layer(TimeoutLayer::new(Duration::from_secs(10)))
.layer(CorsLayer::new().allow_origin(Any));
Router::new()
.route("/health", get(|| async { "ok" }))
.layer(layers)
}async fn shutdown_signal() {
let ctrl_c = async {
tokio::signal::ctrl_c().await.ok();
};
#[cfg(unix)]
let terminate = async {
tokio::signal::unix::signal(tokio::signal::unix::SignalKind::terminate())
.ok()
.and_then(|mut s| s.recv().await);
};
#[cfg(not(unix))]
let terminate = std::future::pending::<()>();
tokio::select! {
_ = ctrl_c => {}
_ = terminate => {}
}
}
#[tokio::main]
async fn main() {
let app = app();
let listener = tokio::net::TcpListener::bind("0.0.0.0:3000").await.unwrap();
axum::serve(listener, app)
.with_graceful_shutdown(shutdown_signal())
.await
.unwrap();
}use axum::{body::Body, http::Request, Router};
use tower::ServiceExt;
#[tokio::test]
async fn health_returns_ok() {
let app: Router = super::app();
let res = app
.oneshot(Request::builder().uri("/health").body(Body::empty()).unwrap())
.await
.unwrap();
assert_eq!(res.status(), 200);
}