1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
use std::sync::{Arc, Mutex};
use serde_json::Value as JsonValue;
use tokio::sync::{broadcast, mpsc};
use super::types::{ConnectionState, McpOrchestratorService};
use harn_serve::transport::{
read_jsonrpc_stdio_frame, write_jsonrpc_stdio_message, JsonRpcStdioFrameStyle,
};
pub(super) async fn run_stdio(service: Arc<McpOrchestratorService>) -> Result<(), String> {
let mut stdin = tokio::io::BufReader::new(tokio::io::stdin());
let mut session = ConnectionState::default();
let mut list_notifications = service.subscribe_list_notifications();
let mut resource_notifications = service.subscribe_resource_notifications();
let mut task_notifications = service.subscribe_task_notifications();
let mut log_notifications = service.subscribe_log_notifications();
let (in_tx, mut in_rx) = mpsc::unbounded_channel();
let input_reader = tokio::spawn(async move {
loop {
match read_jsonrpc_stdio_frame(&mut stdin).await {
Ok(Some(frame)) => {
if in_tx.send(Ok(frame)).is_err() {
break;
}
}
Ok(None) => break,
Err(error) => {
let _ = in_tx.send(Err(error));
break;
}
}
}
});
// Single mpsc fan-in for everything we write to stdout: per-request
// responses, broadcast notifications, and progress updates emitted
// by tool handlers via `harn_vm::mcp_progress`. Funnelling all
// outbound JSON through one writer task means progress lines and
// their final response can never interleave mid-line, and the
// ProgressBus can hand its sender to handle_tools_call without a
// separate channel per request.
let (out_tx, mut out_rx) = mpsc::unbounded_channel::<JsonValue>();
let output_style = Arc::new(Mutex::new(JsonRpcStdioFrameStyle::default()));
let writer_style = output_style.clone();
// Explicit shutdown signal for the writer. We must not couple the
// writer's lifetime to "every `out_tx` clone has been dropped": a
// background task (notably an MCP task-mode tool spawned by
// `create_tool_task`) can hold a progress sender past connection
// shutdown, which would leave `out_rx.recv()` pending forever and wedge
// the awaited `writer` join at EOF — an intermittent, load-dependent
// hang to the nextest slow-test cap (harn#5397). On shutdown we instead
// tell the writer to flush what is already queued and exit.
let (shutdown_tx, mut shutdown_rx) = tokio::sync::oneshot::channel::<()>();
let writer = tokio::spawn(async move {
let mut stdout = tokio::io::stdout();
loop {
tokio::select! {
message = out_rx.recv() => match message {
Some(message) => {
let style = *writer_style.lock().expect("stdio frame style poisoned");
if write_jsonrpc_stdio_message(&mut stdout, &message, style)
.await
.is_err()
{
break;
}
}
None => break,
},
_ = &mut shutdown_rx => {
// Drain anything already queued so a final response is
// never truncated, then exit without waiting on stray
// sender clones.
while let Ok(message) = out_rx.try_recv() {
let style = *writer_style.lock().expect("stdio frame style poisoned");
if write_jsonrpc_stdio_message(&mut stdout, &message, style)
.await
.is_err()
{
break;
}
}
break;
}
}
}
});
let progress_bus = harn_vm::mcp_progress::ProgressBus::from_mpsc(out_tx.clone());
let _bus_guard = harn_vm::mcp_progress::ActiveBusGuard::install(Some(progress_bus));
eprintln!("[harn] MCP stdio server ready");
loop {
tokio::select! {
frame = in_rx.recv() => {
let Some(frame) = frame else {
break;
};
let frame = frame?;
*output_style.lock().expect("stdio frame style poisoned") = frame.style;
let request: JsonValue = match frame.parse_json() {
Ok(value) => value,
Err(_) => continue,
};
let response = service.handle_request(&mut session, request).await;
if !response.is_null() {
let _ = out_tx.send(response);
}
}
notification = list_notifications.recv() => {
match notification {
Ok(notification) => { let _ = out_tx.send(notification); }
Err(broadcast::error::RecvError::Lagged(_)) => continue,
Err(broadcast::error::RecvError::Closed) => break,
}
}
notification = resource_notifications.recv() => {
match notification {
Ok(notification) if session.subscribed_resources.contains(¬ification.uri) => {
let _ = out_tx.send(notification.message);
}
Ok(_) => continue,
Err(broadcast::error::RecvError::Lagged(_)) => continue,
Err(broadcast::error::RecvError::Closed) => break,
}
}
notification = task_notifications.recv() => {
match notification {
Ok(notification) if notification.owner == session.client_identity => {
let _ = out_tx.send(notification.message);
}
Ok(_) => continue,
Err(broadcast::error::RecvError::Lagged(_)) => continue,
Err(broadcast::error::RecvError::Closed) => break,
}
}
notification = log_notifications.recv() => {
match notification {
Ok(notification) if notification.level >= session.log_level => {
let _ = out_tx.send(notification.message);
}
Ok(_) => continue,
Err(broadcast::error::RecvError::Lagged(_)) => continue,
Err(broadcast::error::RecvError::Closed) => break,
}
}
}
}
// Signal the writer to flush and exit. Dropping the two senders we
// hold here (the loop's clone and the ProgressBus's clone in the
// install guard) is not sufficient on its own, because a detached
// background task can still hold a progress sender; the explicit
// shutdown makes teardown independent of that. See the writer's
// construction above for the full rationale (harn#5397).
drop(_bus_guard);
let _ = shutdown_tx.send(());
drop(out_tx);
let _ = writer.await;
input_reader.abort();
let _ = input_reader.await;
Ok(())
}