Class: MCP::Server::Transports::StreamableHTTPTransport

Inherits:
Transport
  • Object
show all
Defined in:
lib/mcp/server/transports/streamable_http_transport.rb

Defined Under Namespace

Classes: InvalidJsonError

Constant Summary collapse

SSE_HEADERS =

x-accel-buffering: no tells reverse proxies (nginx and friends) not to buffer the response, which the spec asks of every SSE stream: a buffering proxy holds events back instead of delivering them as they are written, and on a long-lived subscriptions/listen stream that also swallows the keepalive frames a dropped peer would otherwise be detected by. The TypeScript and Python SDKs send it on their SSE responses for the same reason.

{
  "content-type" => "text/event-stream",
  "cache-control" => "no-cache",
  "connection" => "keep-alive",
  "x-accel-buffering" => "no",
}.freeze
DEFAULT_SESSION_IDLE_TIMEOUT =

Secure defaults for stateful mode. Without a finite idle timeout, sessions live until an explicit client DELETE, so an unauthenticated initialize flood retains unbounded ServerSession objects until memory is exhausted. These defaults expire idle sessions and cap the concurrent count, like the C# SDK (the only reference SDK that hardens this by default, with a 2h idle timeout and a 10k idle-session count). One difference: at the cap this transport rejects a new initialize with 503 (after reclaiming any already-expired slots), whereas the C# SDK evicts the oldest idle session. Rejecting keeps established sessions stable and avoids evicting a legitimate idle session on an attacker's behalf, at the cost of refusing new sessions while genuinely full. Pass session_idle_timeout: nil to opt out of expiry and max_sessions: nil to opt out of the cap.

1800
DEFAULT_MAX_SESSIONS = 
10_000
DEFAULT_MAX_LISTEN_SUBSCRIPTIONS =

Cap on concurrent subscriptions/listen streams (SEP-2575). Each stream holds an open SSE connection for its lifetime, so without a bound an unauthenticated client can retain unbounded connections, like the session-flood case DEFAULT_MAX_SESSIONS guards. A listen request past the cap is rejected with HTTP 503; pass max_listen_subscriptions: nil to opt out.

1_000
DEFAULT_SERVER_TO_CLIENT_REQUEST_TIMEOUT =

Default deadline in seconds for a server-to-client request (sampling, elicitation, roots/list, ping). The spec asks implementations to bound every sent request so a peer that never answers cannot exhaust the sender's resources; without one, a client that opens a session and simply never replies parks a worker thread for good.

Ten minutes matches the TypeScript SDK, which raises its uniform 60-second request default to 600 seconds for the legs of its legacy input_required shim because they are "human-paced, so the 60s protocol default is wrong". Every request this transport can send is that kind of leg: someone answering an elicitation prompt, or the client's own model producing a sample. (The Python SDK leaves the deadline unset and bounds nothing by default.) Deployments that want a tighter bound pass a smaller value here; a single handler that legitimately waits longer passes timeout:.

https://modelcontextprotocol.io/specification/2025-11-25/basic/lifecycle#timeouts

600
DEFAULT_MAX_REQUEST_BYTES =

Default upper bound on the JSON-RPC request body. handle_post reads the whole body into memory and parses it, so without a cap a single unauthenticated POST can allocate gigabytes and OOM the worker. 4 MiB comfortably fits a typical JSON-RPC request (a 4 MiB JSON string decodes to ~3 MiB of base64 payload); raise max_request_bytes: for unusually large payloads. Matches the TypeScript SDK's 4 MB default.

4 * 1024 * 1024
MAX_JSON_NESTING =

Conservative bound on JSON nesting depth, so a deeply nested body cannot exhaust the stack or amplify parse cost (complements the byte cap).

64
MAX_MODERN_REQUEST_NOTIFICATIONS =

Cap on the notifications buffered for one modern request (SEP-2575). The sink holds them in memory until the handler returns, so a handler that emits notifications in proportion to client-supplied input would otherwise let one request grow memory without bound. Notifications past the cap are not delivered and the notify helpers report false, the same non-delivery degradation they have on every other undeliverable path.

1_000
DEFAULT_LISTEN_KEEPALIVE_INTERVAL =

Interval in seconds between SSE keepalive comment frames on a subscriptions/listen stream. Without them a silently dropped connection holds its slot until the next fan-out write fails, so on a quiet server a dead peer would occupy a max_listen_subscriptions slot indefinitely. The periodic write detects the dead peer and frees the slot. Matches the TypeScript SDK's 15-second default; pass listen_keepalive_interval: nil when an upstream proxy pings the stream.

15
REQUIRED_POST_ACCEPT_TYPES_SSE = 
["application/json", "text/event-stream"].freeze
REQUIRED_POST_ACCEPT_TYPES_JSON = 
["application/json"].freeze
REQUIRED_GET_ACCEPT_TYPES = 
["text/event-stream"].freeze
STREAM_WRITE_ERRORS = 
[IOError, Errno::EPIPE, Errno::ECONNRESET].freeze
SESSION_REAP_INTERVAL = 
60
DEFAULT_LOOPBACK_HOSTS =

Loopback hosts always accepted by DNS rebinding protection. A locally bound MCP server (the canonical pattern) is protected out of the box; non-loopback deployments widen the list via allowed_hosts:.

["127.0.0.1", "::1", "localhost"].freeze
NAME_BEARING_METHODS =

JSON-RPC methods whose target name is mirrored into the Mcp-Name header (SEP-2575).

[Methods::TOOLS_CALL, Methods::RESOURCES_READ, Methods::PROMPTS_GET].freeze
LISTEN_FILTER_FIELDS =

Maps broadcast notification methods to the SubscriptionFilter field that opts in to them on a subscriptions/listen stream (SEP-2575). notifications/resources/updated is matched by URI against resourceSubscriptions instead.

{
  Methods::NOTIFICATIONS_TOOLS_LIST_CHANGED => :toolsListChanged,
  Methods::NOTIFICATIONS_PROMPTS_LIST_CHANGED => :promptsListChanged,
  Methods::NOTIFICATIONS_RESOURCES_LIST_CHANGED => :resourcesListChanged,
}.freeze
MODERN_BAD_REQUEST_CODES =

JSON-RPC error codes that surface as HTTP 400 on the modern path. -32601 maps to 404 (disambiguating an unknown method from a legacy HTTP+SSE 404) and everything else, including internal errors, stays 200, matching the Python SDK's status ladder.

[
  ErrorCodes::HEADER_MISMATCH,
  ErrorCodes::MISSING_REQUIRED_CLIENT_CAPABILITY,
  ErrorCodes::UNSUPPORTED_PROTOCOL_VERSION,
  JsonRpcHandler::ErrorCode::PARSE_ERROR,
  JsonRpcHandler::ErrorCode::INVALID_REQUEST,
  JsonRpcHandler::ErrorCode::INVALID_PARAMS,
].freeze

Instance Method Summary collapse

Methods inherited from Transport

#handle_json_request, #open, #send_response

Constructor Details

#initialize(server, stateless: false, enable_json_response: false, session_idle_timeout: UNSET_IDLE_TIMEOUT, max_sessions: DEFAULT_MAX_SESSIONS, allowed_origins: nil, allowed_hosts: nil, dns_rebinding_protection: true, session_request_validator: nil, max_request_bytes: DEFAULT_MAX_REQUEST_BYTES, max_listen_subscriptions: DEFAULT_MAX_LISTEN_SUBSCRIPTIONS, listen_keepalive_interval: DEFAULT_LISTEN_KEEPALIVE_INTERVAL, server_to_client_request_timeout: DEFAULT_SERVER_TO_CLIENT_REQUEST_TIMEOUT) ⇒ StreamableHTTPTransport

Creates a Streamable HTTP transport that can be mounted as a Rack app.

Parameters:

  • server (MCP::Server) —

    the server whose requests this transport dispatches.

  • stateless (Boolean) (defaults to: false) —

    when true, no session is issued and each POST is self-contained.

  • enable_json_response (Boolean) (defaults to: false) —

    when true, a request is answered with a single JSON object instead of an SSE stream.

  • session_idle_timeout (Numeric, nil) (defaults to: UNSET_IDLE_TIMEOUT) —

    seconds before an idle session is reaped; defaults to DEFAULT_SESSION_IDLE_TIMEOUT (1800) in stateful mode, and an explicit nil disables expiry. Not supported in stateless mode.

  • max_sessions (Integer, nil) (defaults to: DEFAULT_MAX_SESSIONS) —

    cap on the concurrent session count in stateful mode; a new initialize past the cap is rejected with HTTP 503, and nil disables the cap.

  • allowed_origins (Array<String>, nil) (defaults to: nil) —

    extra Origin values accepted in addition to same-origin requests, for DNS rebinding protection.

  • allowed_hosts (Array<String>, nil) (defaults to: nil) —

    extra Host values accepted beyond the loopback defaults (127.0.0.1, ::1, localhost); each entry matches a bare host name (any port) or a full host:port.

  • dns_rebinding_protection (Boolean) (defaults to: true) —

    when true (default), validates the Host and Origin headers to prevent DNS rebinding; pass false when an upstream proxy already validates them.

  • session_request_validator (#call, nil) (defaults to: nil) —

    An optional ->(request, session_id) { true | false } invoked on every non-initialize POST, GET, and DELETE against an existing session (regular requests, notifications, and client responses alike). Returning a falsy value rejects the request with HTTP 403. The SDK issues a random SecureRandom.uuid session ID and otherwise only checks existence/idle-timeout, so binding a session to a user is the deploying application's responsibility (the transport never receives the authenticated identity on its own); this is the seam to enforce ownership and mitigate session poisoning. Without a validator, ownership is not enforced.

  • max_request_bytes (Integer) (defaults to: DEFAULT_MAX_REQUEST_BYTES) —

    upper bound in bytes on a POST request body; larger requests are rejected with HTTP 413. Defaults to 4 MiB.

  • max_listen_subscriptions (Integer, nil) (defaults to: DEFAULT_MAX_LISTEN_SUBSCRIPTIONS) —

    cap on concurrent subscriptions/listen streams; a listen request past the cap is rejected with HTTP 503, and nil disables the cap.

  • listen_keepalive_interval (Numeric, nil) (defaults to: DEFAULT_LISTEN_KEEPALIVE_INTERVAL) —

    seconds between SSE keepalive comment frames on a subscriptions/listen stream; the periodic write frees the stream's slot when the peer has gone away. Defaults to DEFAULT_LISTEN_KEEPALIVE_INTERVAL (15); pass nil to disable when an upstream proxy already keeps the stream alive.

  • server_to_client_request_timeout (Numeric) (defaults to: DEFAULT_SERVER_TO_CLIENT_REQUEST_TIMEOUT) —

    seconds a server-to-client request waits for its response before the transport stops waiting and raises MCP::Server::RequestTimeoutError. Defaults to DEFAULT_SERVER_TO_CLIENT_REQUEST_TIMEOUT (600); individual calls override it with timeout:.



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
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
 
# File 'lib/mcp/server/transports/streamable_http_transport.rb', line 135

def initialize(
  server,
  stateless: false,
  enable_json_response: false,
  session_idle_timeout: UNSET_IDLE_TIMEOUT,
  max_sessions: DEFAULT_MAX_SESSIONS,
  allowed_origins: nil,
  allowed_hosts: nil,
  dns_rebinding_protection: true,
  session_request_validator: nil,
  max_request_bytes: DEFAULT_MAX_REQUEST_BYTES,
  max_listen_subscriptions: DEFAULT_MAX_LISTEN_SUBSCRIPTIONS,
  listen_keepalive_interval: DEFAULT_LISTEN_KEEPALIVE_INTERVAL,
  server_to_client_request_timeout: DEFAULT_SERVER_TO_CLIENT_REQUEST_TIMEOUT
)
  super(server)
  # Maps `session_id` to `{ get_sse_stream: stream_object, server_session: ServerSession, last_active_at: float_from_monotonic_clock, origin: origin_header }`.
  @sessions = {}
  @mutex = Mutex.new

  @stateless = stateless
  @enable_json_response = enable_json_response
  @session_request_validator = session_request_validator
  @dns_rebinding_protection = dns_rebinding_protection

  # Host names are case-insensitive, so the allow lists are compared down-cased.
  @allowed_hosts = (DEFAULT_LOOPBACK_HOSTS + Array(allowed_hosts)).map(&:downcase).freeze
  @allowed_origins = Array(allowed_origins).map(&:downcase).freeze
  @pending_responses = {}

  # Maps a `subscriptions/listen` request id to `{ stream: stream_object, filter: honored_subscription_filter }` (SEP-2575).
  # In-process only; a multi-worker deployment needs an external event bus to fan notifications out across processes,
  # which is a follow-up.
  @listen_subscriptions = {}

  # Maps a modern request's ephemeral session id to the Array collecting the notifications its handler emits;
  # `handle_modern` registers the sink and flushes it as SSE frames ahead of the final response (SEP-2575).
  @modern_request_sinks = {}

  # Resolve the idle timeout: an explicit value (including `nil` to opt out) wins; otherwise apply the secure default,
  # which does not apply to stateless mode since it retains no sessions.
  @session_idle_timeout = if session_idle_timeout.equal?(UNSET_IDLE_TIMEOUT)
    stateless ? nil : DEFAULT_SESSION_IDLE_TIMEOUT
  else
    session_idle_timeout
  end

  if @session_idle_timeout
    if @stateless
      raise ArgumentError, "session_idle_timeout is not supported in stateless mode."
    elsif @session_idle_timeout <= 0
      raise ArgumentError, "session_idle_timeout must be a positive number."
    end
  end

  unless max_sessions.nil? || (max_sessions.is_a?(Integer) && max_sessions > 0)
    raise ArgumentError, "max_sessions must be a positive Integer or nil"
  end

  # The cap guards the stateful session store; stateless mode keeps none.
  @max_sessions = stateless ? nil : max_sessions

  unless max_request_bytes.is_a?(Integer) && max_request_bytes > 0
    raise ArgumentError, "max_request_bytes must be a positive Integer"
  end

  @max_request_bytes = max_request_bytes

  if !max_listen_subscriptions.nil? && !(max_listen_subscriptions.is_a?(Integer) && max_listen_subscriptions > 0)
    raise ArgumentError, "max_listen_subscriptions must be a positive Integer or nil"
  end

  @max_listen_subscriptions = max_listen_subscriptions

  if !listen_keepalive_interval.nil? && !(listen_keepalive_interval.is_a?(Numeric) && listen_keepalive_interval > 0)
    raise ArgumentError, "listen_keepalive_interval must be a positive number or nil"
  end

  @listen_keepalive_interval = listen_keepalive_interval

  unless server_to_client_request_timeout.is_a?(Numeric) && server_to_client_request_timeout.positive?
    raise ArgumentError, "server_to_client_request_timeout must be a positive number"
  end

  @server_to_client_request_timeout = server_to_client_request_timeout

  start_reaper_thread if @session_idle_timeout
end

Instance Method Details

#call(env) ⇒ Object

Rack app interface. This transport can be mounted as a Rack app.



259
260
261
 
# File 'lib/mcp/server/transports/streamable_http_transport.rb', line 259

def call(env)
  handle_request(Rack::Request.new(env))
end

#cancel_pending_request(request_id, reason: nil) ⇒ Object

Unblocks a send_request awaiting a response when the peer is being cancelled. The waiting thread will see :cancelled on its queue and raise MCP::CancelledError.

Race note: this is first-writer-wins on the pending-response queue. If a real response has already been pushed (client responded before the cancel hook fired), that response wins and :cancelled is enqueued behind it but never read - send_request returns the real response and deletes the pending entry in its ensure block. Conversely, if :cancelled arrives first, any later client response is silently dropped in handle_response because the pending entry has been removed.



613
614
615
616
617
618
619
620
 
# File 'lib/mcp/server/transports/streamable_http_transport.rb', line 613

def cancel_pending_request(request_id, reason: nil)
  @mutex.synchronize do
    if (pending = @pending_responses[request_id])
      pending[:cancel_reason] = reason
      pending[:queue].push(:cancelled)
    end
  end
end

#close ⇒ Object 



320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
 
# File 'lib/mcp/server/transports/streamable_http_transport.rb', line 320

def close
  @reaper_thread&.kill
  @reaper_thread = nil

  teardown_listen_subscriptions

  removed_sessions = @mutex.synchronize do
    @sessions.each_key.filter_map { |session_id| cleanup_session_unsafe(session_id) }
  end

  removed_sessions.each do |session|
    close_stream_safely(session[:get_sse_stream])
    close_post_request_streams(session)
  end
end

#close_streams(streams) ⇒ Object 



464
465
466
467
468
 
# File 'lib/mcp/server/transports/streamable_http_transport.rb', line 464

def close_streams(streams)
  streams.each do |stream|
    close_stream_safely(stream)
  end
end

#deliver_broadcast_notification(notification) ⇒ Object 



402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
 
# File 'lib/mcp/server/transports/streamable_http_transport.rb', line 402

def deliver_broadcast_notification(notification)
  streams_to_close = []

  # Snapshot the connected streams under the lock, then write to each outside it.
  targets = @mutex.synchronize do
    expired_session_ids = []

    collected = @sessions.filter_map do |session_id, session|
      next unless (stream = session[:get_sse_stream])

      if session_expired?(session)
        expired_session_ids << session_id
        next
      end

      [session_id, stream]
    end

    expired_session_ids.each do |session_id|
      cleanup_and_collect_stream(session_id, streams_to_close)
    end

    collected
  end

  close_streams(streams_to_close)

  targets.count do |session_id, stream|
    write_notification(stream, notification, session_id, nil)
  end
end

#deliver_targeted_notification(notification, session_id, related_request_id) ⇒ Object 



374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
 
# File 'lib/mcp/server/transports/streamable_http_transport.rb', line 374

def deliver_targeted_notification(notification, session_id, related_request_id)
  # JSON response mode returns a single JSON object as the POST response,
  # so request-scoped notifications (e.g. progress, log) cannot be delivered
  # alongside it. Session-scoped standalone notifications
  # (e.g. `resources/updated`, `elicitation/complete`) still flow via GET SSE.
  return false if @enable_json_response && related_request_id

  streams_to_close = []

  # Resolve the target stream under the lock, then write outside it: a stalled SSE reader
  # must not block every other session that needs `@mutex`.
  stream = @mutex.synchronize do
    next unless (session = @sessions[session_id])

    if session_expired?(session)
      cleanup_and_collect_stream(session_id, streams_to_close)
      next
    end

    active_stream(session, related_request_id: related_request_id)
  end

  close_streams(streams_to_close)
  return false unless stream

  write_notification(stream, notification, session_id, related_request_id)
end

#drop_broken_stream(session_id, stream, related_request_id) ⇒ Object

Removes a stream that failed to accept a write. A request-scoped stream is dropped on its own; a session-scoped (GET SSE) failure tears down the whole session. The @sessions mutation runs under @mutex, and the affected streams are closed outside it.



448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
 
# File 'lib/mcp/server/transports/streamable_http_transport.rb', line 448

def drop_broken_stream(session_id, stream, related_request_id)
  streams_to_close = []

  @mutex.synchronize do
    session = @sessions[session_id]
    if related_request_id && session&.dig(:post_request_streams, related_request_id)
      session[:post_request_streams].delete(related_request_id)
      streams_to_close << stream
    else
      cleanup_and_collect_stream(session_id, streams_to_close)
    end
  end

  close_streams(streams_to_close)
end

#handle_request(request) ⇒ Object 



269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
 
# File 'lib/mcp/server/transports/streamable_http_transport.rb', line 269

def handle_request(request)
  rebinding_error = validate_dns_rebinding(request)
  return rebinding_error if rebinding_error

  # Header-primary era routing (SEP-2575). An `MCP-Protocol-Version` header naming a version outside
  # every supported list routes to the sessionless modern path, so an unknown future version receives
  # the spec-mandated `-32022` with the supported list instead of the legacy path's generic invalid-request error.
  # Requests without the header, or with a stable-only version, take the existing paths untouched.
  # An empty header value is malformed rather than a version claim, so it stays on the legacy path
  # and fails legacy header validation as before.
  #
  # A modern header value (2026-07-28) alone cannot decide the era: sessionless modern traffic carries it,
  # and requests of an established legacy session may stamp it as well (the handshake itself never negotiates it):
  # an `Mcp-Session-Id` binds the request to an established legacy session (POST requests, the GET SSE stream,
  # and DELETE termination keep working), and a session-less POST whose body is `initialize` is
  # the legacy-distinctive handshake. Everything else under a dual-era header is sessionless modern traffic
  # (`server/discover`, envelope-carrying requests, and envelope-missing requests that get the modern path's error shape).
  header_version = request.env["HTTP_MCP_PROTOCOL_VERSION"]
  if header_version && !header_version.empty? && !stable_only_version?(header_version)
    unless MCP::Configuration.modern_protocol_version?(header_version)
      return handle_modern(request, header_version)
    end

    if extract_session_id(request).nil?
      return handle_modern(request, header_version) unless request.env["REQUEST_METHOD"] == "POST"

      # The body is readable only once (Rack 3 inputs need not be rewindable), so the era sniff reads it here,
      # bounded, and hands the string to whichever path serves the request.
      body_string = read_bounded_body(request)
      return payload_too_large_response if body_string.nil?

      unless legacy_handshake_body?(body_string)
        return handle_modern(request, header_version, body_string: body_string)
      end

      return handle_post(request, body_string: body_string)
    end
  end

  case request.env["REQUEST_METHOD"]
  when "POST"
    handle_post(request)
  when "GET"
    handle_get(request)
  when "DELETE"
    handle_delete(request)
  else
    method_not_allowed_response
  end
end

#send_notification(method, params = nil, session_id: nil, related_request_id: nil) ⇒ Object 



336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
 
# File 'lib/mcp/server/transports/streamable_http_transport.rb', line 336

def send_notification(method, params = nil, session_id: nil, related_request_id: nil)
  # `subscriptions/listen` streams (SEP-2575) receive matching change notifications regardless of the delivery below:
  # a resource updated by one session's tool call changed globally, so modern subscribers hear about it too.
  # Runs before the per-request sink and the stateless guard because the listen registry does not depend on sessions,
  # and a sink capturing the notification for its own response stream must not hide it from other subscriptions.
  deliver_to_listen_subscriptions(method, params)

  notification = {
    jsonrpc: "2.0",
    method: method,
  }
  notification[:params] = params if params

  # A modern request's notifications ride its own response stream (SEP-2575): `handle_modern` registers
  # a per-request sink for its ephemeral session and flushes it as SSE frames ahead of the final response.
  # Checked before the stateless guard, since modern requests are served in stateless deployments too.
  # The sink is bounded; past `MAX_MODERN_REQUEST_NOTIFICATIONS` the notification is dropped as
  # non-delivery (`false`), never handed to the legacy paths below.
  sink = @mutex.synchronize { session_id && @modern_request_sinks[session_id] }
  if sink
    return false if sink.size >= MAX_MODERN_REQUEST_NOTIFICATIONS

    sink << notification
    return true
  end

  # Stateless mode has no streams to deliver notifications on. Report non-delivery instead of raising
  # so the ephemeral per-request session's notify_* helpers (e.g. progress or log notifications from
  # a tool handler) degrade gracefully rather than spamming the exception reporter on every call.
  return false if @stateless

  if session_id
    deliver_targeted_notification(notification, session_id, related_request_id)
  else
    deliver_broadcast_notification(notification)
  end
end

#send_request(method, params = nil, session_id: nil, related_request_id: nil, parent_cancellation: nil, server_session: nil, timeout: nil) ⇒ Object

Sends a server-to-client JSON-RPC request (e.g., sampling/createMessage) and blocks until the client responds.

Uses a PendingResponse for cross-thread synchronization: this method registers one, sends the request via SSE stream, then waits on it. When the client POSTs a response, handle_response matches it by request_id and resolves the pending response, unblocking this thread. A cancellation and session teardown resolve it the same way.

The wait is bounded by timeout (defaulting to the transport's server_to_client_request_timeout), so a client that never answers cannot park the calling thread for good. On expiry the peer is sent notifications/cancelled and MCP::Server::RequestTimeoutError is raised.



481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
 
# File 'lib/mcp/server/transports/streamable_http_transport.rb', line 481

def send_request(method, params = nil, session_id: nil, related_request_id: nil, parent_cancellation: nil, server_session: nil, timeout: nil)
  # The modern lifecycle (SEP-2575) forbids server-initiated JSON-RPC requests;
  # multi round-trip `input_required` results (SEP-2322) replace them. A modern session has never reached
  # the rest of this method anyway, since `handle_modern` mints its session without registering it in `@sessions`,
  # but that is incidental: without the rule stated here a modern handler that reaches for elicitation
  # or sampling is told "Session not found: <uuid>", which points at everything except the actual reason.
  # `StdioTransport#send_request` refuses the same way.
  if server_session&.era == :modern
    raise "Server-initiated requests are not available in the modern lifecycle (SEP-2575)."
  end

  if @stateless
    raise "Stateless mode does not support server-to-client requests."
  end

  if @enable_json_response
    raise "JSON response mode does not support server-to-client requests."
  end

  unless session_id
    raise "session_id is required for server-to-client requests."
  end

  request_id = generate_request_id
  pending_response = PendingResponse.new
  wait_timeout = timeout || @server_to_client_request_timeout
  cancel_hook = nil

  request = { jsonrpc: "2.0", id: request_id, method: method }
  request[:params] = params if params

  # Register the pending response and resolve the stream under the lock, but perform
  # the write outside it so a stalled reader cannot block every other session on `@mutex`.
  stream = @mutex.synchronize do
    unless (session = @sessions[session_id])
      raise "Session not found: #{session_id}."
    end

    @pending_responses[request_id] = { queue: pending_response, session_id: session_id }

    active_stream(session, related_request_id: related_request_id)
  end

  sent = false
  if stream
    begin
      send_to_stream(stream, request)
      sent = true
    rescue *STREAM_WRITE_ERRORS
      drop_broken_stream(session_id, stream, related_request_id)
    end
  end

  # TODO: Replace with event store + replay when resumability is implemented.
  # Resumability is a separate MCP specification feature (SSE event IDs, Last-Event-ID replay,
  # event store management) independent of sampling.
  # See: https://modelcontextprotocol.io/specification/latest/basic/transports#resumability-and-redelivery
  #
  # The TypeScript and Python SDKs buffer messages and replay on reconnect.
  # Until then, raise to prevent queue.pop from blocking indefinitely.
  unless sent
    raise "No active stream for #{method} request."
  end

  if parent_cancellation && server_session
    cancel_hook = parent_cancellation.on_cancel do |reason|
      server_session.send_peer_cancellation(
        nested_request_id: request_id,
        related_request_id: related_request_id,
        reason: reason,
      )
    end
  end

  response = pending_response.pop(timeout: wait_timeout) do
    # Expiry cancels as well as stops waiting, so a client that answers late does not act on
    # a request the server has abandoned. Only connections speaking 2025-11-25 or earlier get here:
    # the modern lifecycle forbids server-to-client requests outright, and its sessionless requests
    # never register the session this method looks up. Those revisions ask the sender to "issue
    # a cancellation notification for that request and stop waiting", letting either side send one.
    # (The 2026-07-28 rule reserving `notifications/cancelled` for `subscriptions/listen` teardown
    # governs the era that has no such requests to cancel.) Both reference SDKs send this same
    # courtesy cancel on timeout.
    server_session&.send_peer_cancellation(
      nested_request_id: request_id,
      related_request_id: related_request_id,
      reason: "Timed out after #{wait_timeout} seconds",
    )

    raise RequestTimeoutError.new(
      "#{method} request timed out after #{wait_timeout} seconds",
      request_id: request_id,
      timeout: wait_timeout,
    )
  end

  if response.is_a?(Hash) && response.key?(:error)
    raise StandardError, "Client returned an error for #{method} request (code: #{response[:error][:code]}): #{response[:error][:message]}"
  end

  if response == :session_closed
    raise "SSE session closed while waiting for #{method} response."
  end

  if response == :cancelled
    reason = @mutex.synchronize { @pending_responses.dig(request_id, :cancel_reason) }
    raise MCP::CancelledError.new(
      "#{method} request was cancelled",
      request_id: request_id,
      reason: reason,
    )
  end

  response
ensure
  parent_cancellation.off_cancel(cancel_hook) if cancel_hook
  if request_id
    @mutex.synchronize do
      @pending_responses.delete(request_id)
    end
  end
end

#serves_subscriptions_listen? ⇒ Boolean

The subscriptions/listen notification stream (SEP-2575) is served on the modern path, so Server#discover may advertise listChanged/subscribe capability flags.

Returns:

  • (Boolean) 


265
266
267
 
# File 'lib/mcp/server/transports/streamable_http_transport.rb', line 265

def serves_subscriptions_listen?
  true
end

#write_notification(stream, notification, session_id, related_request_id) ⇒ Object

Writes a notification to an SSE stream without holding @mutex. On a write error, drops the broken stream and returns false; on success returns true.



436
437
438
439
440
441
442
443
 
# File 'lib/mcp/server/transports/streamable_http_transport.rb', line 436

def write_notification(stream, notification, session_id, related_request_id)
  send_to_stream(stream, notification)
  true
rescue *STREAM_WRITE_ERRORS => e
  MCP.configuration.exception_reporter.call(e, { session_id: session_id, error: "Failed to send notification" })
  drop_broken_stream(session_id, stream, related_request_id)
  false
end