[Capability] Harden structuredContent extraction (#417)
* Address review feedback on resource_link Three follow-ups from review of the resource_link work: * PromptResultFormatter: add a regression test proving the optional fields of a typed resource_link block (title, description, mimeType, size, annotations, _meta) survive formatting. The delegation to ResourceLink::fromArray() already landed while rebasing onto the PromptResultFormatter refactor, but the existing test only supplied uri/name and so could not catch a reintroduced field drop. * ResourceLink::fromArray(): validate optional fields consistently, the way the equivalent ResourceDefinition::fromArray() already does. description, mimeType and size now raise InvalidArgumentException instead of surfacing a TypeError (or silently coercing, for size's (int) cast); annotations goes through Annotations::tryFromArray(), which carries the is_array() guard that icons already had; icons uses Icon::listFromArray() so a non-array entry is reported in context. * ToolReference::extractStructuredContent(): never emit a list as structuredContent. The Content guard added earlier fixed only one instance of the real invariant - structuredContent must be a JSON object, and a PHP list can never be one. Tool::fromArray() already enforces the matching rule by rejecting an outputSchema whose type is not "object", so this aligns the runtime path with it. The test asserting the opposite for an array-typed outputSchema contradicted the phpstan type and the fromArray() check introduced alongside it in the same commit, so it is replaced by tests for the two list shapes. * Apply the structuredContent object rule to object results too `extractStructuredContent()` guards raw array results against being emitted as a JSON array, but the object branch handed back whatever `json_decode()` produced. A `JsonSerializable` returning a list or a scalar slipped straight through, producing exactly the `structuredContent` the array guard exists to prevent — and a return value that contradicts the method's own `array<string, mixed>|null` signature. Check the decoded value before returning it, and cover the object branch, which had no tests at all. * Document structured tool output `outputSchema` and `structuredContent` were undocumented: the tool return value docs covered only the `content` side, and the schema generation section is about tool parameters. Add a "Structured Output" subsection covering how to declare the schema, which return values populate `structuredContent`, and why a list has to be wrapped in a key to get structured output at all. * Gate structuredContent on the negotiated protocol revision SEP-2106 (revision 2026-07-28) widens structuredContent to any JSON value; earlier revisions require an object. Resolve the revision per request and apply the matching rule, and warn when a declared outputSchema yields none. * Resolve the protocol revision on RequestContext Keeps the `_meta`-then-session lookup in one place instead of per handler, and exposes the negotiated revision to tool handlers. * Document reading the negotiated revision from RequestContext * Warn when a self-built CallToolResult carries an invalid structuredContent Returning a CallToolResult opts out of the extraction rules, so the value is still sent unchanged — but a JSON array is not valid before SEP-2106.
This commit is contained in:
parent
82aac972c1
commit
becce29c41
5 changed files with 616 additions and 36 deletions
|
|
@ -13,6 +13,7 @@ namespace Mcp\Capability\Registry;
|
|||
|
||||
use Mcp\Capability\Formatter\ToolResultFormatter;
|
||||
use Mcp\Schema\Content\Content;
|
||||
use Mcp\Schema\Enum\ProtocolVersion;
|
||||
use Mcp\Schema\Tool;
|
||||
|
||||
/**
|
||||
|
|
@ -59,22 +60,39 @@ class ToolReference extends ElementReference
|
|||
/**
|
||||
* Extracts structured content from a tool result using the output schema.
|
||||
*
|
||||
* @param mixed $toolExecutionResult the raw value returned by the tool's PHP method
|
||||
* What may be sent as `structuredContent` depends on the protocol revision in
|
||||
* use. Up to `2025-11-25` it has to be a JSON object, and `outputSchema` is
|
||||
* restricted to `type: "object"` to match. From `2026-07-28` on (SEP-2106)
|
||||
* `outputSchema` is any JSON Schema 2020-12 and `structuredContent` is any JSON
|
||||
* value conforming to it — a list included.
|
||||
*
|
||||
* @return array<string, mixed>|null the structured content, or null if not extractable
|
||||
* @param mixed $toolExecutionResult the raw value returned by the tool's PHP method
|
||||
* @param ?ProtocolVersion $protocolVersion revision the result is produced for; defaults to the
|
||||
* newest handshake revision, whose stricter rule is what
|
||||
* every revision reachable through `initialize` requires
|
||||
*
|
||||
* @return array<array-key, mixed>|null the structured content, or null if not extractable
|
||||
*
|
||||
* @throws \JsonException if JSON encoding fails for non-Content array/object results
|
||||
*/
|
||||
public function extractStructuredContent(mixed $toolExecutionResult): ?array
|
||||
public function extractStructuredContent(mixed $toolExecutionResult, ?ProtocolVersion $protocolVersion = null): ?array
|
||||
{
|
||||
$objectOnly = ($protocolVersion ?? ProtocolVersion::latestHandshake())->requiresObjectStructuredContent();
|
||||
|
||||
if (\is_array($toolExecutionResult)) {
|
||||
// A PHP list serializes to a JSON array, which the revisions predating
|
||||
// SEP-2106 do not allow as `structuredContent` — strict clients reject
|
||||
// the whole tool call when one is sent.
|
||||
if ($objectOnly && array_is_list($toolExecutionResult)) {
|
||||
return null;
|
||||
}
|
||||
|
||||
foreach ($toolExecutionResult as $item) {
|
||||
if ($item instanceof Content) {
|
||||
// Content items are already reflected in the result's `content`
|
||||
// array; a raw array holding one or more of them isn't
|
||||
// structured data and, if it were serialized as-is, could
|
||||
// produce a `structuredContent` value that isn't a JSON object
|
||||
// (e.g. a list), which the spec doesn't allow.
|
||||
// array; an array holding one or more of them isn't structured
|
||||
// data. This holds in every revision — it is a duplication rule,
|
||||
// not a shape rule.
|
||||
return null;
|
||||
}
|
||||
}
|
||||
|
|
@ -88,9 +106,23 @@ class ToolReference extends ElementReference
|
|||
\JSON_PRETTY_PRINT | \JSON_UNESCAPED_SLASHES | \JSON_UNESCAPED_UNICODE | \JSON_THROW_ON_ERROR | \JSON_INVALID_UTF8_SUBSTITUTE
|
||||
);
|
||||
|
||||
return json_decode(
|
||||
$decoded = json_decode(
|
||||
$jsonResult, true, 512, \JSON_THROW_ON_ERROR
|
||||
);
|
||||
|
||||
// A plain object always encodes to a JSON object, but `JsonSerializable`
|
||||
// can hand back anything. A scalar is dropped whatever the revision
|
||||
// allows: `CallToolResult::$structuredContent` is typed `?array` and
|
||||
// cannot carry one.
|
||||
if (!\is_array($decoded)) {
|
||||
return null;
|
||||
}
|
||||
|
||||
if ($objectOnly && array_is_list($decoded)) {
|
||||
return null;
|
||||
}
|
||||
|
||||
return $decoded;
|
||||
}
|
||||
|
||||
return null;
|
||||
|
|
|
|||
|
|
@ -84,6 +84,15 @@ class ResourceLink extends Content
|
|||
if (isset($data['_meta']) && !\is_array($data['_meta'])) {
|
||||
throw new InvalidArgumentException('Invalid "_meta" in ResourceLink data.');
|
||||
}
|
||||
if (isset($data['description']) && !\is_string($data['description'])) {
|
||||
throw new InvalidArgumentException('Invalid "description" in ResourceLink data.');
|
||||
}
|
||||
if (isset($data['mimeType']) && !\is_string($data['mimeType'])) {
|
||||
throw new InvalidArgumentException('Invalid "mimeType" in ResourceLink data.');
|
||||
}
|
||||
if (isset($data['size']) && !\is_int($data['size'])) {
|
||||
throw new InvalidArgumentException('Invalid "size" in ResourceLink data; expected an integer.');
|
||||
}
|
||||
|
||||
return new self(
|
||||
uri: $data['uri'],
|
||||
|
|
@ -91,9 +100,9 @@ class ResourceLink extends Content
|
|||
title: isset($data['title']) && \is_string($data['title']) ? $data['title'] : null,
|
||||
description: $data['description'] ?? null,
|
||||
mimeType: $data['mimeType'] ?? null,
|
||||
annotations: isset($data['annotations']) ? Annotations::fromArray($data['annotations']) : null,
|
||||
size: isset($data['size']) ? (int) $data['size'] : null,
|
||||
icons: isset($data['icons']) && \is_array($data['icons']) ? array_map(Icon::fromArray(...), $data['icons']) : null,
|
||||
annotations: Annotations::tryFromArray($data['annotations'] ?? null, 'ResourceLink'),
|
||||
size: $data['size'] ?? null,
|
||||
icons: isset($data['icons']) && \is_array($data['icons']) ? Icon::listFromArray($data['icons'], 'ResourceLink') : null,
|
||||
meta: $data['_meta'] ?? null,
|
||||
);
|
||||
}
|
||||
|
|
|
|||
|
|
@ -101,6 +101,20 @@ enum ProtocolVersion: string
|
|||
return $this->isAtLeast(self::FIRST_MODERN_VERSION);
|
||||
}
|
||||
|
||||
/**
|
||||
* Whether this revision restricts `structuredContent` to a JSON object.
|
||||
*
|
||||
* SEP-2106, part of {@see self::V2026_07_28}, widened `outputSchema` to any
|
||||
* JSON Schema 2020-12 and `structuredContent` to any JSON value conforming to
|
||||
* it. Up to `2025-11-25` both are restricted to an object.
|
||||
*
|
||||
* @see https://modelcontextprotocol.io/specification/2026-07-28/server/tools#structured-content
|
||||
*/
|
||||
public function requiresObjectStructuredContent(): bool
|
||||
{
|
||||
return !$this->isAtLeast(self::V2026_07_28);
|
||||
}
|
||||
|
||||
/**
|
||||
* Whether this revision is at least as new as $minimum.
|
||||
*/
|
||||
|
|
|
|||
|
|
@ -22,6 +22,7 @@ use Mcp\Schema\JsonRpc\Request;
|
|||
use Mcp\Schema\JsonRpc\Response;
|
||||
use Mcp\Schema\Request\CallToolRequest;
|
||||
use Mcp\Schema\Result\CallToolResult;
|
||||
use Mcp\Server\RequestContext;
|
||||
use Mcp\Server\Session\SessionInterface;
|
||||
use Psr\Log\LoggerInterface;
|
||||
use Psr\Log\NullLogger;
|
||||
|
|
@ -92,13 +93,37 @@ final class CallToolHandler implements RequestHandlerInterface
|
|||
$arguments['_session'] = $session;
|
||||
$arguments['_request'] = $request;
|
||||
|
||||
$context = new RequestContext($session, $request);
|
||||
|
||||
try {
|
||||
$result = $this->referenceHandler->handle($reference, $arguments);
|
||||
|
||||
$protocolVersion = $context->getProtocolVersion();
|
||||
|
||||
$structuredContent = null;
|
||||
if (!$result instanceof CallToolResult) {
|
||||
$structuredContent = $reference->extractStructuredContent($result);
|
||||
$structuredContent = $reference->extractStructuredContent($result, $protocolVersion);
|
||||
|
||||
if (null === $structuredContent && null !== $reference->tool->outputSchema) {
|
||||
$this->logger->warning('Tool declares an "outputSchema" but returned a value that cannot be sent as "structuredContent"; the value is only carried in "content".', [
|
||||
'name' => $toolName,
|
||||
'result_type' => get_debug_type($result),
|
||||
]);
|
||||
}
|
||||
|
||||
$result = new CallToolResult($reference->formatResult($result), structuredContent: $structuredContent);
|
||||
} elseif ($protocolVersion->requiresObjectStructuredContent()
|
||||
&& \is_array($result->structuredContent)
|
||||
&& [] !== $result->structuredContent
|
||||
&& array_is_list($result->structuredContent)
|
||||
) {
|
||||
// A tool building its own `CallToolResult` bypasses the extraction
|
||||
// rules on purpose, so the value is sent as it was set — but a JSON
|
||||
// array is not valid here before SEP-2106 and clients may reject it.
|
||||
$this->logger->warning('Tool returned a "CallToolResult" whose "structuredContent" is a JSON array, which the negotiated protocol revision does not allow; sending it unchanged.', [
|
||||
'name' => $toolName,
|
||||
'protocol_version' => $protocolVersion->value,
|
||||
]);
|
||||
}
|
||||
|
||||
$this->logger->debug('Tool executed successfully', [
|
||||
|
|
|
|||
|
|
@ -12,6 +12,7 @@
|
|||
namespace Mcp\Server;
|
||||
|
||||
use Mcp\Capability\Logger\ClientLogger;
|
||||
use Mcp\Schema\Enum\ProtocolVersion;
|
||||
use Mcp\Schema\JsonRpc\Request;
|
||||
use Mcp\Server\Session\SessionInterface;
|
||||
|
||||
|
|
@ -25,6 +26,14 @@ use Mcp\Server\Session\SessionInterface;
|
|||
*/
|
||||
final class RequestContext
|
||||
{
|
||||
/**
|
||||
* `_meta` key carrying the protocol revision of a single request, introduced
|
||||
* with the modern era that replaced the `initialize` handshake.
|
||||
*
|
||||
* @see https://modelcontextprotocol.io/specification/2026-07-28/basic/versioning
|
||||
*/
|
||||
private const PROTOCOL_VERSION_META_KEY = 'io.modelcontextprotocol/protocolVersion';
|
||||
|
||||
private ?ClientGateway $clientGateway = null;
|
||||
private ?ClientLogger $clientLogger = null;
|
||||
|
||||
|
|
@ -44,6 +53,26 @@ final class RequestContext
|
|||
return $this->session;
|
||||
}
|
||||
|
||||
/**
|
||||
* The protocol revision this request is served under.
|
||||
*
|
||||
* Modern revisions declare it per request in `_meta`, handshake ones negotiate
|
||||
* it once and keep it on the session. Neither is guaranteed to be present — a
|
||||
* transport may skip `initialize` entirely — so this falls back to the newest
|
||||
* handshake revision, whose rules hold for every revision below it too.
|
||||
*/
|
||||
public function getProtocolVersion(): ProtocolVersion
|
||||
{
|
||||
$requested = $this->request->getMeta()[self::PROTOCOL_VERSION_META_KEY]
|
||||
?? $this->session->get('protocol_version');
|
||||
|
||||
if (!\is_string($requested)) {
|
||||
return ProtocolVersion::latestHandshake();
|
||||
}
|
||||
|
||||
return ProtocolVersion::tryFrom($requested) ?? ProtocolVersion::latestHandshake();
|
||||
}
|
||||
|
||||
public function getClientGateway(): ClientGateway
|
||||
{
|
||||
if (null == $this->clientGateway) {
|
||||
|
|
|
|||
Loading…
Reference in a new issue