Run long work as tasks
阅读约需 99分钟 分钟,8节中含4个示例。
这部分尚未翻译,因此以英文显示。
By the end of this chapter you can negotiate a Model Context Protocol extension, run a slow restock report as a task that the client polls, let tasks expire and cancel them, and say what MCP Apps add to a tool. What the chapter states is true of the specification at revision 2026-07-28 and of mcp 2.3.0, the Python SDK.
Before the shop places its weekly order for kitchen and garden goods with its suppliers, a lead on its support team asks the host for a restock report: every product recounted, and the ones with fewer than 30 units across both warehouses listed for reordering. That is more than a count. The shop's API has a slow endpoint, POST /stock/{sku}/recount, which recounts one product in both warehouses and by default takes 3 seconds. In Serve over Streamable HTTP, the shop's recount_stock tool walks all 64 products through that endpoint and streams a progress report after each one, while the client keeps the HTTP response open.
At the API's default speed, 64 products take more than three minutes, and a call that long has to survive the host's timeout, every proxy between the host and the server, and a lead who closes a laptop. Chapter 6 showed, in Cancel a call the client no longer needs, that a Streamable HTTP client cancels a call by closing its stream, so a call whose connection drops takes its work with it.
Work like this fits another shape that the Model Context Protocol (MCP) offers. With the Tasks extension, the server answers the call at once with a handle to a task: the call's work, whose status the client asks for as often as the server suggests, and whose result waits on the server until the client fetches it. Tasks is an extension, an optional addition to the protocol that both sides must agree to use. This chapter negotiates one, runs the restock report as a task, and ends with when not to.
Every example serves the shop's MCP server over loopback Streamable HTTP, as chapter 6's do, with the shop's API beside it. Setup they share lives in examples/lib/, and two modules there are new. lib.tasks is the book's own small Tasks extension, which Negotiate an extension introduces. lib.restock holds the restock report's work: it recounts every product through chapter 6's lib.recount, sets the task's status message after each one, and lists the products to reorder.
这部分尚未翻译,因此以英文显示。
Negotiate an extension
The core protocol is what every client and server must implement. An extension is something the protocol's maintainers, or anyone else, want to standardize without making it mandatory. Extensions became part of the protocol at revision 2026-07-28, when ClientCapabilities and ServerCapabilities gained an extensions field (changelog). The field maps each extension's identifier to a settings object, and an empty one means the extension is supported with no settings. An extension's identifier is a vendor prefix, a slash and a name. Official extensions use the prefix io.modelcontextprotocol, and anyone else uses a reversed domain they own (Extensions overview).
At the pinned revision, a client declares its extensions in every request, under _meta["io.modelcontextprotocol/clientCapabilities"].extensions, the same place it declares every other capability (Carry the version and capabilities on every request). A server lists its own in the result of server/discover (Extensions overview). Both sides leave extensions off by default and need an explicit opt-in. When only one side supports an extension, the specification says that side "MUST either revert to core protocol behavior or reject the request with an appropriate error" (versioning). So an extension is negotiated, not forked: a server that uses one still answers clients that don't.
Extensions also have a lifecycle. A policy adopted in September 2026, after the pinned revision, gives each extension a stage, Experimental, Beta or Stable, or a path into the core protocol (SEP-3392). Tasks is at Beta, to Core, and MCP Apps, which Know what MCP Apps add covers, is Stable (Extensions overview). A Beta extension may still change in ways that break clients.
The Tasks extension's identifier is io.modelcontextprotocol/tasks, and it defines no settings (SEP-2663). The Python SDK, mcp 2.3.0, ships the framework for writing extensions, but not the Tasks extension itself. A server-side extension subclasses mcp.server.extension.Extension, sets identifier, and can add tools, resources and new request methods, or wrap tools/call. A client-side extension is a ClientExtension, which declares the identifier and can claim result shapes that the core protocol doesn't have (mcp/server/extension.py, mcp/client/extension.py).
The SDK has no tasks/get handler and no types for the extension's handle. The task types it does define belong to the experimental tasks of revision 2025-11-25 (mcp_types). The extension's only official SDK package, @modelcontextprotocol/ext-tasks, is TypeScript, and for the pinned revision it covers only the requester, the client side; its receiver support is for the 2025-11-25 tasks. No official package implements the extension's server side at the pinned revision (ext-tasks).
So this book builds a small one. lib.tasks implements, on the SDK's framework, the part of the extension this chapter teaches: a Tasks server extension, a TasksClient client extension, and the methods tasks/get and tasks/cancel. It leaves out tasks/update and pushed notifications, and it keeps tasks in one process's memory. Use it to learn the protocol, and replace it with the SDK's own when a release ships one.
The shop's server passes tasks.Tasks({"restock_report"}) to MCPServer(extensions=[...]), which tells it to answer every call to restock_report with a task. The tool itself calls lib.restock. Serving the shop's server to one client with TasksClient and one without shows both halves of the negotiation; the example reads the first client's declaration off the wire, and the second client also sends one tasks/get.
1"""Declare the Tasks extension on the shop's server, then show what a client gets with it and without it: both declarations, and the2Missing Required Client Capability error a client without it gets from the call and from the task method. The prose's point rests on those two errors."""3import asyncio4import json5import os67from mcp import Client8from mcp.server.mcpserver import MCPServer9from mcp.shared.exceptions import MCPError1011from lib import loopback, restock, shop_api, tasks, wire1213# only restock_report answers with a task; the server's other tools answer as usual14server = MCPServer("shop-support", version="0.7.0", extensions=[tasks.Tasks({"restock_report"})])151617@server.tool(title="Restock report")18async def restock_report(below: int = 30) -> restock.RestockReport:19 """Recount every product's stock and list those with fewer than `below` units. Slow: runs as a task."""20 return await restock.run(below)212223async def main() -> None:24 app = wire.record_http(server.streamable_http_app(), log="declare.jsonl")25 with loopback.serve(shop_api.create_app()) as api_url, loopback.serve(app) as base_url:26 os.environ.update(shop_api.server_env(api_url))27 async with Client(base_url + "/mcp", extensions=[tasks.TasksClient()]) as client:28 print("protocol revision:", client.protocol_version)29 print("server declares:", json.dumps(client.server_capabilities.extensions))30 await client.list_tools()31 async with Client(base_url + "/mcp") as plain:32 try:33 await plain.call_tool("restock_report")34 except MCPError as error:35 print("client without it:", error.error.code, error.error.message)36 # the specification's field, which lists the extensions the client lacked37 print("required:", json.dumps(error.error.data["requiredCapabilities"]))38 try:39 await tasks.get(plain.session, "any-task-id") # the task methods check the declaration too40 except MCPError as error:41 print("tasks/get without it:", error.error.code)42 # the first tools/list the client sent, as the wire trace recorded it43 sent = next(e["request"] for e in wire.read_http_trace("declare.jsonl") if e["request"]["method"] == "tools/list")44 # at this revision a client declares its capabilities in the _meta of each request it sends45 print("client declares:", json.dumps(sent["params"]["_meta"]["io.modelcontextprotocol/clientCapabilities"]))464748asyncio.run(main())
python declare-the-tasks-extension.pyprotocol revision: 2026-07-28
server declares: {"io.modelcontextprotocol/tasks": {}}
client without it: -32021 Client did not declare required extension 'io.modelcontextprotocol/tasks'
required: {"extensions": {"io.modelcontextprotocol/tasks": {}}}
tasks/get without it: -32021
client declares: {"extensions": {"io.modelcontextprotocol/tasks": {}}}
The server declared {"io.modelcontextprotocol/tasks": {}}, and the client with the extension sent {"extensions": {"io.modelcontextprotocol/tasks": {}}} in the _meta of its tools/list request. Both are the identifier with an empty settings object.
The client without the extension called restock_report and got a JSON-RPC error: -32021 Client did not declare required extension 'io.modelcontextprotocol/tasks'. The error's data lists what that client lacked, {"extensions": {"io.modelcontextprotocol/tasks": {}}}. This is the specification's Missing Required Client Capability error, and the SDK's require_client_extension() raises it with the requiredCapabilities field filled in (mcp/server/mcpserver/server.py). The task methods check too. The same client's tasks/get got -32021, because the specification says servers MUST return that error for tasks/get, tasks/update and tasks/cancel from clients that didn't declare the extension (SEP-2663). The SDK doesn't make that check for an extension's own methods, so lib.tasks calls require_client_extension() in each of them.
The shop rejects the call rather than falling back, because the fallback, a plain three-minute call, is the problem this chapter started from. A tool whose work is sometimes short could fall back instead, and answer clients without the extension the ordinary way.
Working rule: Turn an extension on only on a side that implements it, through the SDK's extension classes, never by writing the identifier into capabilities by hand. Decide, for each tool on the server, what a client without the extension gets: an ordinary result when the work can finish within a normal call, and error -32021 when it can't. Check the client's declaration on every request, because at the pinned revision each request declares its own.
这部分尚未翻译,因此以英文显示。
Start a task instead of waiting
The client in Negotiate an extension declared that it can handle a task, and declaring is all the client does. The server alone decides, call by call and at the pinned revision, whether to answer with a task. A client that declared the extension "MUST be prepared to handle either CallToolResult or CreateTaskResult" (SEP-2663). The same tool can answer one call at once and the next with a task, and nothing in its definition says which. In the experimental tasks of revision 2025-11-25, the client asked for a task with a task parameter on the request. That design is gone and not wire-compatible with the extension.
Three rules limit the server. It MUST NOT return a task to a client that didn't declare the extension on that request, even if the client declared it on an earlier one. Only tools/call can be answered with a task in this revision. And the server MUST NOT return the handle until the task is durably created, so that a tasks/get for it would already find it (SEP-2663).
A task's handle is a CreateTaskResult: a result whose resultType is "task" instead of "complete". It carries the task's taskId, its status, an optional statusMessage, the timestamps createdAt and lastUpdatedAt, its lifetime in ttlMs, and a suggested pollIntervalMs (Tasks schema). The specification lets extensions add resultType values, and a client MUST treat a value it doesn't recognize as invalid (overview). That rule is why a server never sends a task to a client that hasn't declared the extension.
lib.tasks puts all of this in Tasks.intercept_tool_call, the SDK's one hook around tools/call. For a call to restock_report, it checks the request's declaration, creates a task record with an ID from secrets.token_urlsafe(24), and stores it. Then it starts the tool's ordinary handler in the background and returns the handle. Because the background work runs the same handler as a plain call, the task's result is exactly the CallToolResult a plain call returns. Timestamps come from tasks.Clock, a fake clock that starts on the shop's reference date, 30 June 2026, and moves only when an example moves it, so the output is the same on every run.
This run reuses the server from the previous example and calls restock_report with allow_claimed=True, which returns the handle, not the finished report. It then sends one tasks/get straight after, to check that the server already holds the task.
1"""Call restock_report and keep the handle the server answers with at once, instead of waiting for the report. The tasks/get sent straight after is the line the prose rests on: it finds the task, so the server stored it before the handle left."""2import asyncio3import os45from mcp import Client67from lib import http_mcp, loopback, shop_api, tasks, wire89server = http_mcp.load_server("declare-the-tasks-extension.py") # the shop's server with the Tasks extension101112async def main() -> None:13 app = wire.record_http(server.streamable_http_app(), log="start.jsonl")14 with loopback.serve(shop_api.create_app()) as api_url, loopback.serve(app) as base_url:15 os.environ.update(shop_api.server_env(api_url))16 async with Client(base_url + "/mcp", extensions=[tasks.TasksClient()]) as client:17 # allow_claimed hands back the handle itself, not the result the client would otherwise wait for18 handle = await client.session.call_tool("restock_report", {"below": 30}, allow_claimed=True)19 task = await tasks.get(client.session, handle.task_id)20 print("result type:", handle.result_type)21 print("handle fields:", ", ".join(sorted(handle.model_dump(by_alias=True, exclude_none=True))))22 print("status:", handle.status)23 print("created at:", handle.created_at)24 print("ttl ms:", handle.ttl_ms)25 print("poll interval ms:", handle.poll_interval_ms)26 print("task id length:", len(handle.task_id))27 print("tasks/get straight after:", task["status"])28 exchanges = wire.read_http_trace("start.jsonl")29 get = next(e for e in exchanges if e["request"]["method"] == "tasks/get") # the tasks/get the client sent above30 # the trace stores header names in lowercase, so the lookup uses mcp-name31 assert get["request_headers"].get("mcp-name") == handle.task_id323334asyncio.run(main())
python start-a-restock-report.pyresult type: task
handle fields: _meta, createdAt, lastUpdatedAt, pollIntervalMs, resultType, status, taskId, ttlMs
status: working
created at: 2026-06-30T09:00:00Z
ttl ms: 3600000
poll interval ms: 100
task id length: 32
tasks/get straight after: working
The call came back at once as a task, not the report. Its resultType was task, and its status was working. The handle's fields were _meta, createdAt, lastUpdatedAt, pollIntervalMs, resultType, status, taskId, ttlMs. The _meta field is the SDK's serverInfo stamp, added to every result at the pinned revision (mcp/server/runner.py).
The task was created at 2026-06-30T09:00:00Z on the fake clock. Its ttlMs was 3600000, one hour, and its pollIntervalMs was 100. The tasks/get sent straight after found the task working, which shows the server stored it before the handle left.
The task ID was 32 characters long, which encodes 192 random bits. The specification says task IDs may act as bearer handles, so servers MUST generate them with sufficient entropy that a third party cannot enumerate or guess them (SEP-2663). Over Streamable HTTP, the client MUST also copy the ID into the Mcp-Name header of tasks/get, tasks/update and tasks/cancel, so that a load balancer can route every request for one task the same way. The client's tasks/get carried the task ID in Mcp-Name, because lib.tasks sets name_param = "taskId" on its request classes and the SDK's client fills the header from that field (mcp_types/_types.py).
Working rule: Answer with a task when the work can outlive the connection, and decide per call, not per tool, when some calls are short. Store the task before you return the handle, and generate its ID from a cryptographic random source, with at least 128 bits. Run the same code a plain call runs, so a task's result never differs from a plain result. Never return a task to a request that didn't declare the extension.
这部分尚未翻译,因此以英文显示。
Poll a task and fetch its result
The handle from Start a task instead of waiting said the restock report was working. A task has five statuses: working, input_required, completed, failed and cancelled. The completed, failed and cancelled statuses are terminal, and once a task reaches one, its state doesn't change (Tasks overview). failed means a JSON-RPC error. A tool that ran and returned a result with isError: true ends completed, with that result inside, and the specification says failed MUST NOT be used for it (SEP-2663). It is the same split chapter 3 drew between tool execution errors and protocol errors, in Report failures as tool results.
The client finds out where a task stands by polling. It sends tasks/get with the taskId, and it SHOULD wait the server's pollIntervalMs between polls, a value that may change from one reply to the next. Servers MAY rate-limit clients that poll faster. Each reply is the task, with resultType "complete". When the task has completed, the reply carries result, which has the same structure as the result type of the original request (SEP-2663; Tasks schema). For a tool, that is a CallToolResult. Clients SHOULD store task IDs durably, so that polling can resume after a crash or a restart.
A server may also push status changes. The client subscribes with subscriptions/listen for the task IDs it holds, and the server sends notifications/tasks, each carrying the full task. Polling stays the default (Tasks overview), and lib.tasks implements only polling. A task that needs the user's input moves to input_required and carries input requests shaped like a Multi Round-Trip Request's (Return an input request instead of a result), which the client answers with tasks/update. lib.tasks doesn't implement that either, and the restock report never asks.
In the SDK, polling belongs in a client extension. TasksClient claims the result type "task" with a ResultClaim. The claim's resolver gets the handle and a session for follow-up requests, and it must return the call's ordinary CallToolResult (mcp/client/extension.py). The resolver waits pollIntervalMs, sends tasks/get, and repeats until the task is terminal. Client.call_tool then returns the result as if the call had been plain. A client without a claim for "task" can't parse the handle: the SDK validates it as a CallToolResult and raises pydantic's ValidationError. UnexpectedClaimedResult, which carries the handle, is for a client that has the claim but calls session.call_tool without allow_claimed=True (mcp/client/session.py).
A host makes this call when a lead asks for the restock report. Through TasksClient, with on_poll= set to keep every tasks/get reply, a client calls restock_report with a threshold of 30 units.
1"""Call restock_report through a client whose extension polls the task until it ends, and get the report back as an ordinary tool result. The asserts check that the report was still working on more than one poll and that its messages only counted up, which the prose relies on."""2import asyncio3import os45from mcp import Client67from lib import http_mcp, loopback, shop_api, tasks89server = http_mcp.load_server("declare-the-tasks-extension.py") # the shop's server with the Tasks extension101112async def main() -> None:13 polls = []14 with loopback.serve(shop_api.create_app()) as api_url, loopback.serve(server.streamable_http_app()) as base_url:15 os.environ.update(shop_api.server_env(api_url))16 # on_poll keeps every tasks/get reply, in the order the client received them17 async with Client(base_url + "/mcp", extensions=[tasks.TasksClient(on_poll=polls.append)]) as client:18 result = await client.call_tool("restock_report", {"below": 30})19 statuses = [p["status"] for p in polls]20 # dict.fromkeys drops repeated statuses but keeps their order, so each status is listed once21 print("statuses:", " then ".join(dict.fromkeys(statuses)))22 assert statuses.count("working") > 1, statuses23 # the second word of "recounted N of 64 SKUs" is N24 counted = [int(p["statusMessage"].split()[1]) for p in polls if "statusMessage" in p]25 assert counted == sorted(counted), counted26 print("last message:", polls[-1]["statusMessage"])27 print("result type:", type(result).__name__)28 report = result.structured_content29 print("skus recounted:", report["skus"])30 print("to reorder:", len(report["reorder"]))31 # the first of the products the report lists for reordering32 print("first to reorder:", "{sku}, {units} units".format(**report["reorder"][0]))333435asyncio.run(main())
python poll-the-restock-report.pystatuses: working then completed
last message: recounted 64 of 64 SKUs
result type: CallToolResult
skus recounted: 64
to reorder: 5
first to reorder: KG-1018, 2 units
The polls show the statuses in order: working then completed. It was still working on more than one poll, its status messages only counted up, and the last one read "recounted 64 of 64 SKUs". The number of polls varies with the machine's speed, so the example prints none.
What came back from Client.call_tool was a CallToolResult. The report recounted 64 products and listed 5 to reorder, the first being KG-1018, 2 units. The code that made the call never saw a task, which is the point of the claim: a host written against plain tool calls, such as the one in Build a host that works through tickets, gets tasks by adding one extension to its client.
Working rule: Poll at the server's pollIntervalMs, re-read it on every reply, and stop at the first terminal status. Treat completed as the original call's result, isError and all, and failed as a protocol error. Hide the polling in a client extension so the rest of the host sees plain calls. When a host must survive a restart, store each task ID before the first poll, and resume from it.
这部分尚未翻译,因此以英文显示。
Expire and cancel tasks
The restock report in Poll a task and fetch its result finished and was fetched. Not every task is. A client may crash and never come back, or a lead may close a report they no longer need, and a server that keeps every task forever runs out of storage. The extension gives a task two ways to end early: its lifetime runs out, or the client cancels it.
A task's lifetime is its ttlMs, counted from createdAt, or null for no limit, and the server may change it. After the TTL elapses, servers MAY mark a task failed at any point, and later delete it at any time. A server that has purged a task must answer a tasks/get for it with error -32602, Invalid Params, and it may say the task can't be found rather than that it expired (SEP-2663). So a client can't count on a result outliving the TTL, however the task went.
Cancelling uses tasks/cancel with the taskId, and the specification says notifications/cancelled MUST NOT be used for it. The reply is an empty acknowledgement, and it promises nothing more: "A server is not obligated to actually stop the work; it is only obligated to acknowledge the request." The task may stay working for a while, and it may still end completed or failed. Once it has sent the cancel, the client may forget the task (SEP-2663). This is cooperative cancellation, unlike chapter 6's, where closing the stream cancelled the call's work in the server (Cancel a call the client no longer needs).
lib.tasks makes both concrete. On a tasks/get after the TTL, it deletes the task, stops its work, and answers -32602 with the message the specification's example uses. On tasks/cancel, it sets a flag and acknowledges. The tool's work reads the flag through tasks.cancel_requested(), and lib.restock checks it after each product, so the restock report stops at the end of the product it is recounting, and the task becomes cancelled then.
The expiry example sets a one-minute ttlMs, which shows expiry without a long wait. So that its output never depends on the machine's speed, it holds each report before every product, with restock.Steps, a gate in lib.restock that lets one more product start each time allow() is called. It starts one report that is never allowed a product, polls it, moves the fake clock 61 seconds on, and polls again. Then it starts a second report, lets it recount one product, waits for its first status message, and cancels it. It polls at once, lets the report start its next product, and polls until the task is terminal.
1"""Show the two ways a task ends early: its time to live runs out and the server purges it, or the client cancels it and the report stops at its next check. The -32602 from the purged task and the final status of the cancelled report are the lines the prose rests on."""2import asyncio3import os45from mcp import Client6from mcp.server.mcpserver import MCPServer7from mcp.shared.exceptions import MCPError89from lib import loopback, restock, shop_api, tasks1011clock = tasks.Clock() # fake time: moves only when the example says so12first_steps, second_steps = restock.Steps(), restock.Steps() # a report starts a product only when allowed13pending = [first_steps, second_steps] # one per call, in call order14# a task expires one minute after it was created, counted on the fake clock15server = MCPServer("shop-support", version="0.7.0",16 extensions=[tasks.Tasks({"restock_report"}, ttl_ms=60_000, clock=clock)])171819@server.tool(title="Restock report")20async def restock_report(below: int = 30) -> restock.RestockReport:21 """Recount every product's stock and list those with fewer than `below` units. Slow: runs as a task."""22 # calls arrive in order, so each report takes the next gate23 return await restock.run(below, steps=pending.pop(0))242526async def main() -> None:27 with loopback.serve(shop_api.create_app()) as api_url, loopback.serve(server.streamable_http_app()) as base_url:28 os.environ.update(shop_api.server_env(api_url))29 async with Client(base_url + "/mcp", extensions=[tasks.TasksClient()]) as client:30 first = await client.session.call_tool("restock_report", allow_claimed=True)31 print("first, ttl ms:", first.ttl_ms) # first_steps allows no product, so it stays working32 print("first, first poll:", (await tasks.get(client.session, first.task_id))["status"])33 clock.advance(61) # 61 seconds: past the one-minute time to live34 try:35 await tasks.get(client.session, first.task_id)36 except MCPError as error:37 print("first, 61 s later:", error.error.code, error.error.message)38 second = await client.session.call_tool("restock_report", allow_claimed=True)39 second_steps.allow() # product 1 only: the report then waits before product 240 task = await tasks.poll_until(client.session, second.task_id, lambda t: "statusMessage" in t)41 print("second, before cancel:", task["status"] + ",", task["statusMessage"])42 ack = await tasks.cancel(client.session, second.task_id)43 print("cancel acknowledged, besides _meta:", ", ".join(f"{k} {v}" for k, v in ack.items() if k != "_meta"))44 print("second, right after:", (await tasks.get(client.session, second.task_id))["status"])45 second_steps.allow() # product 2 starts; the check after it sees the cancel46 task = await tasks.poll_until(client.session, second.task_id, lambda t: t["status"] in tasks.TERMINAL)47 print("second, finally:", task["status"] + ",", task["statusMessage"])484950asyncio.run(main())
python expire-a-task.pyfirst, ttl ms: 60000
first, first poll: working
first, 61 s later: -32602 Failed to retrieve task: Task has expired
second, before cancel: working, recounted 1 of 64 SKUs
cancel acknowledged, besides _meta: resultType complete
second, right after: working
second, finally: cancelled, stopped after 2 of 64 SKUs
The first task's ttlMs was 60000, and it was working on its first poll. After 61 seconds on the fake clock, tasks/get answered -32602 Failed to retrieve task: Task has expired. On that poll, lib.tasks stopped the first task's work too, and its result would have gone with it had the work finished.
The second task was working, recounted 1 of 64 SKUs when the client cancelled it. Besides the SDK's _meta stamp, the acknowledgement held only resultType complete: an empty result, apart from the field every result carries. A poll straight after found the task still working, because the cancel only set a flag and the report hadn't reached its next check. In the shop, that check comes after the current product's recount; in the example, the report was held before its second product.
The last poll found cancelled, stopped after 2 of 64 SKUs. The report recounted one more product after the cancel, then stopped, as the specification allows.
Working rule: Give every task a TTL that covers the longest wait you expect between the work finishing and the client fetching it, and purge after it; for the shop's restock report, that is an hour. Make long work check for cancellation at each natural step, such as each product, and clean up there. In the client, treat a cancel's acknowledgement as "asked", not "stopped", and treat -32602 from tasks/get as a task that is gone for good.
这部分尚未翻译,因此以英文显示。
Know what MCP Apps add
Tasks change when a tool's result arrives, as Start a task instead of waiting showed. MCP Apps is an official extension that changes how a result is shown. Its identifier is io.modelcontextprotocol/ui. A tool names a user interface in _meta.ui.resourceUri: a ui:// resource, served with the MIME type text/html;profile=mcp-app. A host that supports Apps renders that HTML in a sandboxed iframe, and talks to it only through postMessage, in JSON-RPC: MCP's own methods, such as tools/call, plus methods whose names start with ui/. Through the host, the app can call the server's tools (MCP Apps; Apps specification).
A tool's _meta.ui.visibility says who may call it, ["model", "app"] by default. A tool for the app alone is hidden from the model. The Apps specification puts both directions on the host: it MUST NOT include such a tool in the model's tool list, and it MUST reject an app's call to a tool whose visibility leaves out "app" (Apps specification). A UI-enabled tool should still return meaningful text for clients that don't support the extension (Extensions overview).
The Apps specification is dated 2026-01-26 and marked Stable, and its negotiation examples use the 2025 initialize handshake. At the pinned revision, a client declares Apps per request, like any extension (Negotiate an extension). Unlike Tasks, Apps ships in the Python SDK, in mcp.server.apps. Apps() is the extension, @apps.tool(resource_uri=...) declares a UI-enabled tool, apps.add_html_resource() serves the HTML, and client_supports_apps(ctx) says whether this request's client declared Apps with the right MIME type (mcp/server/apps.py).
The restock report is a candidate. As an app, it could show the products to reorder as a table, with a button beside each that adds the product to the supplier order. That button would call a tool such as add_to_supplier_order, with a visibility of ["app"], so that the model never places orders on its own. The report would keep its text and structured content, so a host without Apps would read the same report as in Poll a task and fetch its result. The book doesn't build this view, because its host loop in chapter 10 has no screen, and an Apps view needs one to render it.
MCP Apps and elicitation do different jobs. Elicitation is the server asking the user one question in the middle of a call, through the client's own form (Ask for structured input with elicitation). An app is a view of a tool's result, which the server writes and the host renders.
Working rule: Reach for MCP Apps when a result needs more than text to act on, such as a table to sort or a choice per row, and the hosts your users run support the extension. Keep every UI-enabled tool's text result complete, so a host without Apps loses only the view. Give app-only tools a visibility of ["app"] when the model shouldn't call them, and remember that the host enforces it, not your server.
在任务、进度通知与普通调用之间做选择
商店现在以三种方式运行工作:普通调用,例如 get_order_status;流式返回进度的调用,即第 6 章的 recount_stock;以及任务,即 轮询任务并获取其结果 一节中的补货报告。它们的差别在于连接必须撑过什么,以及调用方在等待期间能看到什么。
进度是决定大多数选择的差别。就当前修订版而言,规范并不普遍支持在任务上使用 notifications/progress 或 notifications/message(SEP-2663)。任务只能通过 statusMessage 报告自己进行到了哪里,客户端在轮询时读取它。补货报告中的“已重新清点 64 个 SKU 中的 12 个”就是这类消息:它是一个字符串,客户端轮询多频繁,就只能看到多频繁,也没有进度条可以使用的数值。日志消息也不能绕开这一限制(替换采样、根目录与日志)。
另一个差别在于谁必须支持什么。普通调用和进度属于核心协议。任务则需要双方都实现一个扩展,而宿主对扩展的支持程度不一(扩展支持矩阵)。任务概览列出了适合使用任务的情形:工作的持续时间超过连接的存续时间,工作必须在客户端重启后继续存在,以及工作本来就在队列中作为作业运行。
下表按我们的经验法则(并非实测结果)对三种方式做了比较。
| 普通调用 | 进度 | 任务 | |
|---|---|---|---|
| 典型时长 | 数秒 | 最多几分钟 | 数分钟或更久 |
| 连接必须保持打开 | 是 | 是 | 否 |
| 调用方等待时看到的内容 | 无 | 进度和总量 | 状态消息 |
| 客户端重启后仍然有效 | 否 | 否 | 是,前提是客户端保存了 ID |
| 需要扩展 | 否 | 否 | 是,双方都需要 |
get_order_status 仍然是普通调用,因为它在几毫秒内就能给出答复。recount_stock 沿用第 6 章的进度,因为发起它的组长会看着计数上升并等待。补货报告改为任务,因为按该 API 的实际速度它要运行数分钟,组长希望稍后再取而不是一直盯着看,而且连接中断不应让三分钟的重新清点白费。
工作准则: 当结果能在宿主与服务器之间所有超时限制之内很快返回时,使用普通调用。当工作耗时较长,调用方在一条你信得过会保持打开的连接上等待,并且想要看到一个数值时,使用进度。当工作可能比连接存续得更久,或者调用方想先离开、稍后再回来时,使用任务。要接受任务只能报告一条状态消息,并为没有该扩展的客户端保留普通调用的后备方案,或者明确返回 -32021。
生产环境中会出什么问题
本章的补货报告运行在单个进程中,任务保存在内存里,时钟也是模拟的。若把它部署到多个副本上,连续运行数日,并由许多用户调用,就可能以下列方式出错,每一种都有可供查找的迹象。
任务只存储在一个副本上。lib.tasks 把任务保存在一个字典中。在第 6 章的两个副本之后部署时(在没有粘性会话的情况下运行副本),如果 tasks/get 请求到达另一个副本,就什么也找不到,并返回 -32602。应当把任务存放在所有副本都能读取的地方,例如数据库或作业队列自身的存储;也可以根据 Mcp-Name 标头进行路由,该标头在每个任务请求中都携带任务 ID。其表现为任务在几秒内就“过期”,频率与负载均衡器切换副本的频率相近。
**任务 ID 泄露其他调用方的结果。**任务 ID 可能是调用方与已存储结果之间唯一的屏障。规范指出,服务器必须对每个与任务相关的请求执行认证和授权检查,并且规范移除了 tasks/list,因为它的作用域一旦设置不当,就可能让一个调用方看到另一个调用方的任务 ID(SEP-2663)。应当把调用方与任务一起记录下来,待使用 OAuth 授权客户端就绪后,取自令牌的主体;并拒绝其他任何人发来的每一个 tasks/get 和 tasks/cancel。通过任务到达的输入请求,与直接调用中的输入请求获得同等的信任,不会更多:“任务不是更高信任度的通道。”
**SDK 不会替你检查声明。**在 mcp 2.3.0 中,服务器运行器会不加检查地放行扩展的 resultType,源码中的一条注释说明,对未声明该扩展的客户端拒绝它,这一点仍有待实现(mcp/server/runner.py)。扩展必须自行调用 require_client_extension(),在调用本身以及它自己的每个方法上都要如此,lib.tasks 就是这样做的。未声明 Tasks 却仍然收到任务的客户端,必须把该结果视为无效。SDK 的客户端正是如此:没有针对 "task" 的声明,它会把该句柄当作 CallToolResult 来验证,并抛出 ValidationError。
**轮询的开销超过了工作本身。**成千上万个客户端每 100 毫秒轮询一次,本身就构成一份负载。随着任务运行时间变长,应调大 pollIntervalMs;按规范的允许,对忽略它的客户端进行限流;一旦服务器能够推送,就提供 notifications/tasks。其迹象是在你的追踪记录中,tasks/get 的数量超过其他所有方法。
**任务不断堆积。**TTL 为 null,或者比任何客户端等待的时间都长,就会保留每个被放弃任务的结果。应设置 TTL,除读取时清理外还要按计划定期清理,并对已存储任务的数量设置告警。客户端重启时丢失任务 ID 会让堆积更快增长,因此应在第一次轮询之前就把 ID 存储在客户端。
**已取消的工作仍在运行。**从不检查取消状态的工具,在每次 tasks/cancel 之后仍会运行到结束,而商店的 API 会重新统计没人需要的商品。应在每一步都进行检查,并记录任务在取消之后又运行了多久。
**客户端使用 2025-11-25 版的任务。**该修订版中的实验性任务使用 tasks/result、tasks/list 以及请求上的 task 参数,它们与该扩展在传输格式上并不兼容(SEP-2663)。2025 年代的客户端从不声明 io.modelcontextprotocol/tasks,因此会遇到服务器返回的 -32021。旧协议的其余部分见迁移 2025 年代的服务器。
**扩展会在你不知情时变化。**Tasks 目前为 Beta,正在进入核心协议,SDK 也可能在后续版本中自带 Tasks 扩展。这两种情况都是本章需要更新的触发条件。当 SDK 提供该扩展时,用它替换 lib.tasks,并保持商店的工具不变,因为该工具从来不知道自己是作为任务运行的。
**工作准则:**在任务投入生产之前,把任务的存储迁移到所有副本都能读取的地方;记录每个任务的所有者,并在每个任务请求上检查;设置 TTL,并配以按计划运行的清理。然后按副本监控 tasks/get 返回的 -32602、tasks/get 与 tools/call 的比值、已存储任务的数量,以及取消之后工作还会运行多久。
本章做出的决定
- 商店的补货报告通过 Tasks 扩展以任务方式运行,因为它的工作时间会比一次连接存续得更久。
- 对于没有该扩展的客户端,商店的服务器会以错误 -32021 拒绝
restock_report,而不是运行一次长达三分钟的普通调用。 - 本书在 SDK 的扩展框架上自行构建了一个小型 Tasks 扩展
lib.tasks,因为mcp2.3.0 没有提供这样的扩展。它实现了tasks/get和tasks/cancel,而省去了tasks/update和服务器推送的通知。 - 任务运行的是工具的普通处理函数,因此它的结果与普通调用返回的
CallToolResult相同。 - 任务 ID 含有 192 个随机比特,并且每个任务请求都在
Mcp-Name请求头中携带它的 ID。 - 客户端在一个客户端扩展内部按服务器给出的
pollIntervalMs进行轮询,因此宿主看到的只是普通的工具调用。 - 任务在创建一小时后过期,取消是协作式的:补货报告在处理完每个产品后检查一次。
recount_stock保留第 6 章的进度,get_order_status仍然是普通调用。- MCP Apps 只作介绍而不构建,因为本书的宿主循环不渲染任何 HTML。补货视图会保留报告的文本,并通过将可见性设为
["app"],对模型隐藏补货视图的下单工具。 - 在生产环境中,任务迁移到第 6 章的两个副本都能共享的存储中,并且每个任务请求都要对照启动该任务的调用方进行检查。
- 这一课讲的是等待放在何处:任务把等待从连接上移到服务器上,由服务器保存状态、设定其持续时间,并决定谁可以读取它。
本章,在本版次中
版次 2026.11.14示例于 2026年11月19日 通过本章 4个示例 个示例中有 4 个通过
译自英文版图书,版次 2026.11.14。