Skip to content

runner

ServerRunner - the per-connection handler kernel.

Bridges the dispatch layer (on_request / on_notify, untyped dicts) and the user's handler layer (typed Context, typed params). It holds a pre-populated Connection and a Server (the handler registry); the drivers (serve_stream, serve_connection, serve_one) construct the Connection and tear it down.

aclose_shielded async

aclose_shielded(connection: Connection) -> None

Unwind connection.exit_stack under a shielded, bounded scope.

Called from a driver's finally: the shield lets per-connection cleanup callbacks run even when the driver itself is being cancelled, the _EXIT_STACK_CLOSE_TIMEOUT bound stops a hung callback wedging shutdown, and a raising callback is logged-and-swallowed so it never masks the driver's own exception.

Source code in src/mcp/server/runner.py
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
async def aclose_shielded(connection: Connection) -> None:
    """Unwind ``connection.exit_stack`` under a shielded, bounded scope.

    Called from a driver's ``finally``: the shield lets per-connection cleanup
    callbacks run even when the driver itself is being cancelled, the
    `_EXIT_STACK_CLOSE_TIMEOUT` bound stops a hung callback wedging shutdown,
    and a raising callback is logged-and-swallowed so it never masks the
    driver's own exception.
    """
    with anyio.move_on_after(_EXIT_STACK_CLOSE_TIMEOUT, shield=True) as scope:
        try:
            await connection.exit_stack.aclose()
        except Exception:
            logger.exception("connection exit_stack cleanup raised")
    if scope.cancelled_caught:
        logger.warning(
            "connection exit_stack cleanup exceeded %s seconds; abandoning remaining callbacks",
            _EXIT_STACK_CLOSE_TIMEOUT,
        )

modern_error_data

modern_error_data(exc: Exception) -> ErrorData

Map a modern request's handler exception to its wire ErrorData via the shared handler_exception_to_error_data ladder; anything unmapped is logged and becomes a generic INTERNAL_ERROR so handler internals never reach the wire.

Source code in src/mcp/server/runner.py
462
463
464
465
466
467
468
469
470
def modern_error_data(exc: Exception) -> ErrorData:
    """Map a modern request's handler exception to its wire `ErrorData` via the shared
    `handler_exception_to_error_data` ladder; anything unmapped is logged and becomes a
    generic INTERNAL_ERROR so handler internals never reach the wire."""
    error = handler_exception_to_error_data(exc)
    if error is not None:
        return error
    logger.exception("modern request handler raised")
    return ErrorData(code=INTERNAL_ERROR, message="Internal server error")

modern_on_request

modern_on_request(
    server: Server[LifespanT], lifespan_state: LifespanT
) -> OnRequest

Return an OnRequest callback that serves each call via serve_one with a fresh per-request Connection.

Wire this into the server side of a DirectDispatcher peer-pair (built with can_send_request=False) to drive an in-process server on the modern per-request-envelope path. Like serve_one, raises whatever the handler chain raises.

Source code in src/mcp/server/runner.py
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
def modern_on_request(server: Server[LifespanT], lifespan_state: LifespanT) -> OnRequest:
    """Return an `OnRequest` callback that serves each call via `serve_one` with a fresh per-request `Connection`.

    Wire this into the server side of a `DirectDispatcher` peer-pair (built
    with `can_send_request=False`) to drive an in-process server on the modern
    per-request-envelope path. Like `serve_one`, raises whatever the handler
    chain raises.
    """

    async def handle(
        dctx: DispatchContext[TransportContext], method: str, params: Mapping[str, Any] | None
    ) -> dict[str, Any]:
        meta = (params or {}).get("_meta", {})
        connection = Connection.from_envelope(
            meta.get(PROTOCOL_VERSION_META_KEY, LATEST_MODERN_VERSION),
            meta.get(CLIENT_INFO_META_KEY),
            meta.get(CLIENT_CAPABILITIES_META_KEY),
        )
        return await serve_one(server, dctx, method, params, connection=connection, lifespan_state=lifespan_state)

    return handle

serve_connection async

serve_connection(
    server: Server[LifespanT],
    dispatcher: JSONRPCDispatcher[Any],
    *,
    connection: Connection,
    lifespan_state: LifespanT,
    init_options: InitializationOptions | None = None,
    task_status: TaskStatus[None] = TASK_STATUS_IGNORED
) -> None

Drive dispatcher in handshake mode until the underlying channel closes.

The legacy loop driver over a caller-built dispatcher and Connection; tears down connection.exit_stack (shielded) on the way out. Stream transports use serve_stream; this is for entries that already own the connection's era.

Source code in src/mcp/server/runner.py
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
async def serve_connection(
    server: Server[LifespanT],
    dispatcher: JSONRPCDispatcher[Any],
    *,
    connection: Connection,
    lifespan_state: LifespanT,
    init_options: InitializationOptions | None = None,
    task_status: anyio.abc.TaskStatus[None] = anyio.TASK_STATUS_IGNORED,
) -> None:
    """Drive `dispatcher` in handshake mode until the underlying channel closes.

    The legacy loop driver over a caller-built dispatcher and `Connection`; tears
    down `connection.exit_stack` (shielded) on the way out. Stream transports use
    `serve_stream`; this is for entries that already own the connection's era.
    """
    runner = ServerRunner(server, connection, lifespan_state, init_options=init_options)
    try:
        await dispatcher.run(
            runner.on_request, runner.on_notify, admit=_legacy_admission(runner), task_status=task_status
        )
    finally:
        await aclose_shielded(connection)

serve_one async

serve_one(
    server: Server[LifespanT],
    dctx: DispatchContext[TransportContext],
    method: str,
    params: Mapping[str, Any] | None,
    *,
    connection: Connection,
    lifespan_state: LifespanT
) -> dict[str, Any]

Handle a single request (method, params) and return its result dict.

The single-exchange driver: builds the kernel, runs on_request once under dctx, and tears down connection.exit_stack (shielded) on the way out. The entry constructs the (born-ready) Connection and the dctx; this only consumes them.

Raises whatever the handler chain raises (MCPError / ValidationError / unmapped); callers own the exception-to-wire mapping.

Source code in src/mcp/server/runner.py
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
async def serve_one(
    server: Server[LifespanT],
    dctx: DispatchContext[TransportContext],
    method: str,
    params: Mapping[str, Any] | None,
    *,
    connection: Connection,
    lifespan_state: LifespanT,
) -> dict[str, Any]:
    """Handle a single request ``(method, params)`` and return its result dict.

    The single-exchange driver: builds the kernel, runs `on_request` once under
    `dctx`, and tears down `connection.exit_stack` (shielded) on the way out.
    The entry constructs the (born-ready) `Connection` and the `dctx`; this
    only consumes them.

    Raises whatever the handler chain raises (`MCPError` / `ValidationError` /
    unmapped); callers own the exception-to-wire mapping.
    """
    runner = ServerRunner(server, connection, lifespan_state)
    try:
        return await runner.on_request(dctx, method, params)
    finally:
        await aclose_shielded(connection)

ServerRunner dataclass

Bases: Generic[LifespanT]

Per-connection handler kernel. One instance per client connection.

Source code in src/mcp/server/runner.py
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
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
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
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
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
373
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
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
@dataclass
class ServerRunner(Generic[LifespanT]):
    """Per-connection handler kernel. One instance per client connection."""

    server: Server[LifespanT]
    connection: Connection
    lifespan_state: LifespanT
    _: KW_ONLY
    init_options: InitializationOptions | None = None
    """`InitializeResult` payload. Defaults to `server.create_initialization_options()`."""

    @cached_property
    def on_request(self) -> OnRequest:
        return self._on_request

    @cached_property
    def on_notify(self) -> OnNotify:
        return self._on_notify

    async def _on_request(
        self,
        dctx: DispatchContext[TransportContext],
        method: str,
        params: Mapping[str, Any] | None,
    ) -> dict[str, Any]:
        meta = _extract_meta(params)
        version = self.connection.protocol_version
        ctx = self._make_context(dctx, method, params, meta, version)

        async def _inner(ctx: ServerRequestContext[LifespanT, Any]) -> HandlerResult:
            # Read method/params off `ctx` so a middleware that rewrote them via
            # `call_next(replace(ctx, ...))` reaches lookup and the handler.
            method, params = ctx.method, ctx.params
            # Pinned compat: spec methods are surface-validated before lookup,
            # so malformed params are INVALID_PARAMS even with no handler
            # registered. Custom methods miss the monolith map and fall through
            # to `entry.params_type` exactly as before.
            if method in _methods.SPEC_CLIENT_METHODS:
                try:
                    _methods.validate_client_request(method, version, params)
                except KeyError:
                    raise MCPError(code=METHOD_NOT_FOUND, message="Method not found", data=method) from None
            # TODO(L29): the 2026-07-28 spec drops the handshake; this branch and
            # the gate become a per-version legacy path then. Initialize runs inline
            # (read loop parked), so awaiting the peer anywhere on this path deadlocks.
            if method == "initialize":
                return self._serialize(method, version, self._handle_initialize(params))
            # Methods without a handler are METHOD_NOT_FOUND regardless of
            # initialization state: JSON-RPC 2.0 reserves -32601 for "not
            # available on this server", and clients probing a server before
            # the handshake key off that code. The init gate below therefore
            # only ever applies to methods the server actually serves.
            entry = self.server.get_request_handler(method)
            if entry is None:
                raise MCPError(code=METHOD_NOT_FOUND, message="Method not found", data=method)
            if not self.connection.initialize_accepted and method not in _INIT_EXEMPT:
                # Pinned compat: the same error shape the union validation produced.
                raise MCPError(code=INVALID_PARAMS, message="Invalid request parameters", data="")
            # Absent params validate as {} (required fields still reject), so
            # the handler receives the model with its defaults, never None.
            typed_params = entry.params_type.model_validate({} if params is None else params, by_name=False)
            result = await entry.handler(ctx, typed_params)
            if isinstance(result, ErrorData):
                # Raise inside the chain so middleware observes the failure.
                raise MCPError.from_error_data(result)
            # Shape for the wire inside the chain so the OpenTelemetry span (the
            # outermost middleware) records a failing handler return shape too.
            return self._serialize(method, version, result)

        call = self._compose_server_middleware(_inner)
        # `_inner` already produced the wire dict; a middleware that short-circuited
        # without `call_next` is trusted to return its own well-formed result -
        # including its response envelope. The pipeline never patches it up after
        # the fact.
        result = _dump_result(await call(ctx))
        if method == "initialize":
            # Commit only on chain success, so a middleware veto leaves no state.
            # Race-free: the read loop is parked until this call returns.
            # TODO: this re-reads the wire `params`, so a middleware that rewrote
            # `ctx.params` (or `ctx.method`, or short-circuited without `call_next`)
            # can leave `connection.protocol_version` out of step with the
            # `InitializeResult` `_inner` produced. Resolve when `initialize` becomes
            # a built-in handler so commit and result derive from one negotiation.
            self.connection.client_params, self.connection.protocol_version = self._negotiate_initialize(params)
        return result

    async def _on_notify(
        self,
        dctx: DispatchContext[TransportContext],
        method: str,
        params: Mapping[str, Any] | None,
    ) -> None:
        meta = _extract_meta(params)
        version = self.connection.protocol_version
        ctx = self._make_context(dctx, method, params, meta, version)

        async def _inner(ctx: ServerRequestContext[LifespanT, Any]) -> None:
            method, params = ctx.method, ctx.params
            if method in _methods.SPEC_CLIENT_NOTIFICATION_METHODS:
                try:
                    _methods.validate_client_notification(method, version, params)
                except KeyError:
                    logger.debug("dropped %r: not defined at %s", method, version)
                    return
                except ValidationError:
                    logger.warning("dropped %r: malformed params", method)
                    return
            if method == "notifications/initialized":
                # Surface validation above already rejected a malformed body, so
                # commit; fall through so a registered handler observes an
                # initialized connection.
                self.connection.initialized.set()
            elif not self.connection.initialize_accepted:
                logger.debug("dropped %s: received before initialization", method)
                return
            entry = self.server.get_notification_handler(method)
            if entry is None:
                logger.debug("no handler for notification %s", method)
                return
            # Same absent-params contract as requests.
            try:
                typed_params = entry.params_type.model_validate({} if params is None else params, by_name=False)
            except ValidationError:
                logger.warning("dropped %r: malformed params", method)
                return
            await entry.handler(ctx, typed_params)

        call = self._compose_server_middleware(_inner)
        try:
            await call(ctx)
        except Exception:
            # A crashing handler must not cancel the dispatcher's task group;
            # middleware saw the raise out of call_next() first.
            logger.exception("notification handler for %r raised", method)

    def _compose_server_middleware(self, inner: CallNext) -> CallNext:
        """Wrap `inner` in `Server.middleware`, outermost-first.

        Shared by `_on_request` and `_on_notify` so the same middleware chain
        observes every inbound message. The composed callable takes the `ctx`
        at call time, so a middleware can rewrite it for the rest of the chain.
        """
        call = inner
        for middleware in reversed(self.server.middleware):
            call = partial(_apply_middleware, middleware, call)
        return call

    def _make_context(
        self,
        dctx: DispatchContext[TransportContext],
        method: str,
        params: Mapping[str, Any] | None,
        meta: RequestParamsMeta | None,
        protocol_version: str,
    ) -> ServerRequestContext[LifespanT, Any]:
        # TODO(L54): remove for Context rework. Reads the SHTTP per-request
        # data off the raw `dctx.message_metadata` carrier; replace with the
        # per-transport context once that lands.
        md = dctx.message_metadata
        if isinstance(md, ServerMessageMetadata):
            request = md.request_context
            close_sse_stream = md.close_sse_stream
            close_standalone_sse_stream = md.close_standalone_sse_stream
        else:
            request = close_sse_stream = close_standalone_sse_stream = None
        # Per-request session: `dctx` is the request-scoped channel (auto-threads
        # its own request_id on streamable HTTP); the standalone channel is read
        # off `connection.outbound`. `related_request_id` on the public API selects.
        session = ServerSession(dctx, self.connection)
        return ServerRequestContext(
            session=session,
            lifespan_context=self.lifespan_state,
            method=method,
            params=params,
            request_id=dctx.request_id,
            meta=meta,
            protocol_version=protocol_version,
            request=request,
            close_sse_stream=close_sse_stream,
            close_standalone_sse_stream=close_standalone_sse_stream,
        )

    def _serialize(self, method: str, version: str, result: HandlerResult) -> dict[str, Any]:
        """Shape a handler result into its wire form: the outbound counterpart
        of the inbound classification ladder.

        One pass owns the whole response envelope, in order: cache hints fill
        `ttlMs`/`cacheScope` the handler left unset, core-vocabulary spec-method
        results are validated and sieved by the per-version surface (a claimed
        extension `resultType` shape is the extension's to own), and 2026-era
        results get the `serverInfo` `_meta` stamp (spec #3002). Runs inside the
        middleware chain so the OpenTelemetry span observes a failing return
        shape (unsupported type, malformed spec result) as an error rather
        than closing on a request that the client sees fail - and so a
        middleware that short-circuits without `call_next` owns its result,
        envelope included.
        """
        # MRTR carve-out: `input_required` interim results, typed or mapping, never get hints.
        if (hint := self.server.cache_hints.get(method)) is not None:
            if isinstance(result, CacheableResult):
                result = apply_cache_hint(result, hint)
            elif isinstance(result, Mapping) and not _methods.is_input_required(result):
                # Hint keys first so wire keys the handler set win, matching `apply_cache_hint` precedence.
                result = {"ttlMs": hint.ttl_ms, "cacheScope": hint.scope, **result}
        dumped = _dump_result(result)
        # A modern-era extension `resultType` (outside the core vocabulary) marks
        # a claimed shape owned by the extension that defined it: the per-version
        # surface doesn't describe it, so the sieve applies to core results only.
        # Legacy connections sieve everything - claimed shapes are 2026-era
        # vocabulary and cannot be delivered on a legacy wire (mirrors the
        # client-side ResultClaim rule).
        # TODO(L56): reject extension resultType values unless the corresponding
        # extension is in this request's _meta clientCapabilities.extensions; the
        # explicit MUST-reject is client-side (basic/index.mdx ResultType), this enforces it proactively.
        result_type = dumped.get("resultType")
        core_shape = (
            version not in MODERN_PROTOCOL_VERSIONS
            or not isinstance(result_type, str)
            or result_type in CORE_RESULT_TYPES
        )
        if method in _methods.SPEC_CLIENT_METHODS and core_shape:
            try:
                dumped = _methods.serialize_server_result(method, version, dumped)
            except ValidationError:
                # Server bug, not client fault. Detail stays in the server log:
                # pydantic messages echo the result body.
                logger.exception("handler for %r returned an invalid result", method)
                raise MCPError(code=INTERNAL_ERROR, message="Handler returned an invalid result") from None
        if version in MODERN_PROTOCOL_VERSIONS and dumped.get("resultType") is None:
            # Spec 2026-07-28: `Result.resultType` is required - servers MUST
            # include it (the absent-means-complete bridge is for clients of
            # older servers only). The sieve guarantees it for core methods;
            # this covers everything else: custom methods, extension methods,
            # and empty results.
            dumped["resultType"] = "complete"
        return self._stamp_server_info(version, dumped)

    def _stamp_server_info(self, version: str, result: dict[str, Any]) -> dict[str, Any]:
        """Fill the `serverInfo` `_meta` stamp on a 2026-era result (spec #3002).

        A handler-authored value wins; an explicit `null` reads as absent and
        is stamped over, mirroring the request-side `clientInfo` posture (a
        `null` is not a valid `Implementation`, so presence means a value). A
        non-mapping `_meta` is the handler's to own, and handshake-era results
        are never stamped. `result` is
        pipeline-owned (`_dump_result` copies dicts; the spec-method sieve
        re-dumps), but `_meta` may still be the handler's object, so the stamp
        replaces it rather than writing into it. `server_info_stamp` is a
        fresh dict per access, so the response never aliases server state.
        """
        if version not in MODERN_PROTOCOL_VERSIONS:
            return result
        raw_meta = result.get("_meta")
        if raw_meta is None:
            result["_meta"] = {SERVER_INFO_META_KEY: self.server.server_info_stamp}
        elif isinstance(raw_meta, dict):
            meta = cast("dict[str, Any]", raw_meta)
            if meta.get(SERVER_INFO_META_KEY) is None:
                result["_meta"] = {**meta, SERVER_INFO_META_KEY: self.server.server_info_stamp}
        return result

    @staticmethod
    def _negotiate_initialize(params: Mapping[str, Any] | None) -> tuple[InitializeRequestParams, str]:
        """Validate `initialize` params and pick the protocol version."""
        init = InitializeRequestParams.model_validate(params or {}, by_name=False)
        requested = init.protocol_version
        negotiated = requested if requested in HANDSHAKE_PROTOCOL_VERSIONS else LATEST_HANDSHAKE_VERSION
        return init, negotiated

    def _handle_initialize(self, params: Mapping[str, Any] | None) -> InitializeResult:
        """Build the `initialize` result; state commits later in `_on_request`."""
        _, negotiated = self._negotiate_initialize(params)
        opts = self.init_options if self.init_options is not None else self.server.create_initialization_options()
        return InitializeResult(
            protocol_version=negotiated,
            capabilities=opts.capabilities,
            server_info=Implementation(
                name=opts.server_name,
                title=opts.title,
                description=opts.description,
                version=opts.server_version,
                website_url=opts.website_url,
                icons=opts.icons,
            ),
            instructions=opts.instructions,
        )

init_options class-attribute instance-attribute

init_options: InitializationOptions | None = None

InitializeResult payload. Defaults to server.create_initialization_options().

Classes

  • ServerMiddleware — Context-tier middleware: (ctx, call_next) -> result.
  • ServerRunner — Per-connection handler kernel. One instance per client connection.

Functions

  • aclose_shielded — Unwind connection.exit_stack under a shielded, bounded scope.
  • modern_error_data — Map a modern request's handler exception to its wire ErrorData via the shared handler_exception_to_error_data ladder; anything unmapped is logged and becomes a generic INTERNAL_ERROR so handler internals never reach the wire.
  • modern_on_request — Return an OnRequest callback that serves each call via serve_one with a fresh per-request Connection.
  • serve_connection — Drive dispatcher in handshake mode until the underlying channel closes.
  • serve_one — Handle a single request (method, params) and return its result dict.

Attributes

  • CallNext — Invokes the rest of the chain with the given context. What a context rewrite (dataclasses.replace(ctx, ...)) can alter depends on the tier: ServerMiddleware runs before params validation, so its rewrites change what the handler is invoked with; an Extension interceptor runs after, so its rewrites change only what the handler observes on ctx.