[{"_1":2,"_351":-5,"_352":-5},"loaderData",{"_3":4,"_22":23,"_267":268},"root",{"_5":6,"_7":8},"toasts",[],"visitor",{"_9":10,"_11":12},"viewportWidth",1280,"visitorSession",{"_13":14,"_16":17,"_18":-5,"_19":20,"_21":-5},"attributes",[15],"bot","id",19346472,"membership","referenceId","0199f420-7955-7299-8090-ea86aef37824","userAccount","routes/_public/mcp/servers/~namespace/~slug/_pages/_layout",{"_24":25,"_77":-5,"_78":38,"_27":79,"_182":183,"_265":266},"availableLocales",[26,38,44,50,57,64,71],{"_27":28,"_31":35,"_36":37},"mcpServer",{"_29":30,"_31":32,"_33":34},"description","Bietet Integration zwischen [Genkit](https://github.com/firebase/genkit) und dem Model Context Protocol (MCP).","name","Genkit MCP","readme","# Genkit MCP\n\n> \\[!WARNUNG]\\\n> Dieses Plugin ist experimentell, d. h. es wird möglicherweise nicht langfristig unterstützt und APIs unterliegen häufiger fehlerhaften Änderungen.\n\nDieses Plugin ermöglicht die Integration zwischen Genkit und dem [Model Context Protocol](https://modelcontextprotocol.io) (MCP). MCP ist ein offener Standard, der es Entwicklern ermöglicht, „Server“ zu erstellen, die Tools, Ressourcen und Eingabeaufforderungen für Clients bereitstellen. Genkit MCP ermöglicht es Genkit-Entwicklern, MCP-Tools, -Eingabeaufforderungen und -Ressourcen sowohl als Client zu nutzen als auch Tools und Eingabeaufforderungen als Server bereitzustellen.\n\n## Installation\n\nUm zu beginnen, benötigen Sie Genkit und das MCP-Plugin:\n\n```bash\nnpm i genkit genkitx-mcp\n```\n\n## MCP-Client\n\nUm einen MCP-Client zu erstellen, rufen Sie die Funktion `mcpClient` auf, um ein Genkit-Plugin für einen MCP-Server zu generieren. So verwenden Sie beispielsweise den Beispiel- [Dateisystemserver](https://github.com/modelcontextprotocol/servers/tree/main/src/filesystem) von MCP:\n\n```ts\nimport { genkit } from 'genkit';\nimport { mcpClient } from 'genkitx-mcp';\n\n// the filesystem server requires one or more allowed directories\nconst ALLOWED_DIRS = ['/Users/yourusername/Desktop'];\n\nconst filesystemClient = mcpClient({\n name: 'filesystem',\n serverProcess: {\n command: 'npx',\n args: ['-y', '@modelcontextprotocol/server-everything', ...ALLOWED_DIRS],\n },\n});\n\nconst ai = genkit({\n plugins: [\n filesystemClient /* ... other plugins such as model providers ...*/,\n ],\n});\n```\n\nDie meisten MCP-Server sind so konzipiert, dass sie als gestartete Prozesse auf demselben Rechner über den `stdio` Transport laufen. Mit der Option `serverProcess` geben Sie den Befehl, die Argumente und die Umgebungsvariablen für die Erstellung des Servers als Unterprozess an.\n\n### mcpClient()-Optionen\n\n* **`name`** : (erforderlich) Der Name für diesen Client, der seinen Tools und Eingabeaufforderungen Namespaces zuweist.\n* **`version`** : (optional) Die Versionsnummer des Clients. Der Standardwert ist „1.0.0“.\n* Sie müssen eines der folgenden Elemente angeben:\n * **`serverProcess`** : Parameter zum Starten eines lokalen Serverprozesses mithilfe des stdio-MCP-Transports.\n * **`command`** : Shell-Befehlspfad zum Starten des MCP-Servers. Kann z. B. `npx` oder `uvx` sein, um den Server von einem Paketmanager herunterzuladen und auszuführen.\n * **`args`** : (optional) Array von String-Argumenten, die an den Befehl übergeben werden sollen.\n * **`env`** : (optional) Schlüsselwertobjekt von Umgebungsvariablen, die an den Befehl übergeben werden sollen.\n * **`serverUrl`** : Die URL eines Remote-Servers, zu dem über den SSE MCP-Transport eine Verbindung hergestellt werden soll.\n * \\*\\* `serverWebsocketUrl` : Die URL eines Remote-Servers, zu dem über den WebSocket-MCP-Transport eine Verbindung hergestellt werden soll.\n * **`transport`** : Ein vorhandenes MCP-Transportobjekt für die Verbindung mit dem Server.\n* **`rawToolResponses`** : (optional) Ein Boolean-Flag. Bei `true` werden Tool-Antworten im MCP-Rohformat zurückgegeben; andernfalls werden sie für die Genkit-Kompatibilität verarbeitet.\n\n### Verwenden von MCP-Aktionen\n\nDer Genkit MCP-Client erkennt automatisch verfügbare Tools und Eingabeaufforderungen und registriert sie bei Genkit. Dadurch sind sie überall verfügbar, wo andere Tools und Eingabeaufforderungen verwendet werden können. Für den Zugriff auf Ressourcen werden spezielle `list_resources` und `read_resource` -Tools registriert, die auf Ressourcen für den Server zugreifen.\n\nAlle MCP-Aktionen werden unter dem von Ihnen angegebenen Namen in Namespaces zusammengefasst, sodass ein Client namens `filesystem` Tools wie `filesystem/read_file` registriert.\n\n### Tool-Antworten\n\nMCP-Tools geben ein `content` zurück, im Gegensatz zu einer strukturierten Antwort wie die meisten Genkit-Tools. Das Genkit MCP-Plugin versucht, den zurückgegebenen Inhalt zu analysieren und zu verarbeiten:\n\n1. Wenn der Inhalt Text und gültiges JSON ist, wird das JSON analysiert und zurückgegeben.\n2. Wenn der Inhalt Text und kein gültiges JSON ist, wird der Text zurückgegeben.\n3. Wenn der Inhalt einen einzelnen Nicht-Textteil enthält, wird dieser zurückgegeben.\n4. Wenn der Inhalt aus mehreren/gemischten Teilen besteht, wird die vollständige Inhaltsantwort zurückgegeben.\n\n## MCP-Server\n\nSie können auch alle Tools und Eingabeaufforderungen einer Genkit-Instanz als MCP-Server verfügbar machen:\n\n```ts\nimport { genkit, z } from 'genkit';\nimport { mcpServer } from 'genkitx-mcp';\n\nconst ai = genkit({});\n\nai.defineTool(\n {\n name: 'add',\n description: 'add two numbers together',\n inputSchema: z.object({ a: z.number(), b: z.number() }),\n outputSchema: z.number(),\n },\n async ({ a, b }) => {\n return a + b;\n }\n);\n\nai.definePrompt(\n {\n name: \"happy\",\n description: \"everybody together now\",\n input: {\n schema: z.object({\n action: z.string().default(\"clap your hands\").optional(),\n }),\n },\n },\n `If you're happy and you know it, {{action}}.`\n);\n\nmcpServer(ai, { name: 'example_server', version: '0.0.1' }).start();\n```\n\nDas obige Beispiel startet einen MCP-Server mit dem stdio -Transport, der ein Tool namens `add` und eine Eingabeaufforderung namens `happy` bereitstellt. Um den Server mit einem anderen Transport zu starten, verwenden Sie `mcpServer(...).start(otherTransport)` .\n\n### Bekannte Einschränkungen\n\n* MCP-Eingabeaufforderungen können nur Zeichenfolgenparameter annehmen, daher müssen Eingaben für Schemata Objekte mit ausschließlich Zeichenfolgeneigenschaftswerten sein.\n* MCP-Eingabeaufforderungen unterstützen nur `user` und `model` . `system` werden nicht unterstützt.\n* MCP-Eingabeaufforderungen unterstützen nur einen einzigen „Typ“ innerhalb einer Nachricht, sodass Sie Medien und Text nicht in derselben Nachricht mischen können.\n\n### Testen Ihres MCP-Servers\n\nSie können Ihren MCP-Server mit dem offiziellen Inspector testen. Wenn Ihr Servercode beispielsweise in `dist/index.js` kompiliert wurde, können Sie Folgendes ausführen:\n\n```\nnpx @modelcontextprotocol/inspector dist/index.js\n```\n\nSobald Sie den Inspektor starten, können Sie Eingabeaufforderungen und Aktionen auflisten und manuell testen.","German","tag","de-DE",{"_27":39,"_31":42,"_36":43},{"_29":40,"_31":32,"_33":41},"Provides integration between [Genkit](https://github.com/firebase/genkit) and the Model Context Protocol (MCP).","# Genkit MCP\n\n> [!WARNING] \n> This plugin is experimental, meaning it may not be supported long-term and APIs are subject to more often breaking changes.\n\nThis plugin provides integration between Genkit and the [Model Context Protocol](https://modelcontextprotocol.io) (MCP). MCP is an open standard allowing developers to build \"servers\" which provide tools, resources, and prompts to clients. Genkit MCP allows Genkit developers to both consume MCP tools, prompts, and resources as a client and provide tools and prompts as a server.\n\n## Installation\n\nTo get started, you'll need Genkit and the MCP plugin:\n\n```bash\nnpm i genkit genkitx-mcp\n```\n\n## MCP Client\n\nTo create an MCP client, you call the `mcpClient` function to generate a Genkit plugin for an MCP server. For example, to use MCP's example [filesystem server](https://github.com/modelcontextprotocol/servers/tree/main/src/filesystem):\n\n```ts\nimport { genkit } from 'genkit';\nimport { mcpClient } from 'genkitx-mcp';\n\n// the filesystem server requires one or more allowed directories\nconst ALLOWED_DIRS = ['/Users/yourusername/Desktop'];\n\nconst filesystemClient = mcpClient({\n name: 'filesystem',\n serverProcess: {\n command: 'npx',\n args: ['-y', '@modelcontextprotocol/server-everything', ...ALLOWED_DIRS],\n },\n});\n\nconst ai = genkit({\n plugins: [\n filesystemClient /* ... other plugins such as model providers ...*/,\n ],\n});\n```\n\nMost MCP servers are built to run as spawned processes on the same machine using the `stdio` transport. When you supply the `serverProcess` option, you are specifying the command, arguments, and environment variables for spawning the server as a subprocess.\n\n### mcpClient() Options\n\n- **`name`**: (required) The name for this client, which namespaces its tools and prompts.\n- **`version`**: (optional) The client's version number. Defaults to \"1.0.0\".\n- You must supply one of:\n - **`serverProcess`**: Parameters for launching a local server process using the stdio MCP transport.\n - **`command`**: Shell command path for launching the MCP server. Can be e.g. `npx` or `uvx` to download and run the server from a package manager.\n - **`args`**: (optional) Array of string arguments to pass to the command.\n - **`env`**: (optional) Key value object of environment variables to pass to the command.\n - **`serverUrl`**: The URL of a remote server to connect to using the SSE MCP transport.\n - **`serverWebsocketUrl`: The URL of a remote server to connect to using the WebSocket MCP transport.\n - **`transport`**: An existing MCP transport object for connecting to the server.\n- **`rawToolResponses`**: (optional) A boolean flag. If `true`, tool responses are returned in their raw MCP format; otherwise, they are processed for Genkit compatibility.\n\n### Using MCP Actions\n\nThe Genkit MCP client automatically discovers available tools and prompts and registers them with Genkit making them available anywhere other tools and prompts can be used. To access resources, special `list_resources` and `read_resource` tools are registered that will access resources for the server.\n\nAll MCP actions are namespaced under the name you supply, so a client called `filesystem` will register tools such as `filesystem/read_file`.\n\n### Tool Responses\n\nMCP tools return a `content` array as opposed to a structured response like most Genkit tools. The Genkit MCP plugin attempts to parse and coerce returned content:\n\n1. If content is text and valid JSON, the JSON is parsed and returned.\n2. If content is text and not valid JSON, the text is returned.\n3. If content has a single non-text part, it is returned.\n4. If content has multiple/mixed parts, the full content response is returned.\n\n## MCP Server\n\nYou can also expose all of the tools and prompts from a Genkit instance as an MCP server:\n\n```ts\nimport { genkit, z } from 'genkit';\nimport { mcpServer } from 'genkitx-mcp';\n\nconst ai = genkit({});\n\nai.defineTool(\n {\n name: 'add',\n description: 'add two numbers together',\n inputSchema: z.object({ a: z.number(), b: z.number() }),\n outputSchema: z.number(),\n },\n async ({ a, b }) => {\n return a + b;\n }\n);\n\nai.definePrompt(\n {\n name: \"happy\",\n description: \"everybody together now\",\n input: {\n schema: z.object({\n action: z.string().default(\"clap your hands\").optional(),\n }),\n },\n },\n `If you're happy and you know it, {{action}}.`\n);\n\nmcpServer(ai, { name: 'example_server', version: '0.0.1' }).start();\n```\n\nThe above will start up an MCP server with the stdio transport that exposes a tool called `add` and a prompt called `happy`. To start the server with a different transport, use `mcpServer(...).start(otherTransport)`.\n\n### Known Limitations\n\n- MCP prompts are only able to take string parameters, so inputs to schemas must be objects with only string property values.\n- MCP prompts only support `user` and `model` messages. `system` messages are not supported.\n- MCP prompts only support a single \"type\" within a message so you can't mix media and text in the same message.\n\n### Testing your MCP server\n\nYou can test your MCP server using the official inspector. For example, if your server code compiled into `dist/index.js`, you could run:\n\n npx @modelcontextprotocol/inspector dist/index.js\n\nOnce you start the inspector, you can list prompts and actions and test them out manually.\n","English","en-US",{"_27":45,"_31":48,"_36":49},{"_29":46,"_31":32,"_33":47},"Proporciona integración entre [Genkit](https://github.com/firebase/genkit) y el Protocolo de Contexto de Modelo (MCP).","# Genkit MCP\n\n> \\[!ADVERTENCIA]\\\n> Este complemento es experimental, lo que significa que es posible que no tenga soporte a largo plazo y que las API estén sujetas a cambios disruptivos con mayor frecuencia.\n\nEste complemento proporciona integración entre Genkit y el [Protocolo de Contexto de Modelo](https://modelcontextprotocol.io) (MCP). MCP es un estándar abierto que permite a los desarrolladores crear servidores que proporcionan herramientas, recursos y avisos a los clientes. Genkit MCP permite a los desarrolladores consumir herramientas, avisos y recursos de MCP como cliente y proporcionar herramientas y avisos como servidor.\n\n## Instalación\n\nPara comenzar, necesitará Genkit y el complemento MCP:\n\n```bash\nnpm i genkit genkitx-mcp\n```\n\n## Cliente MCP\n\nPara crear un cliente MCP, se llama a la función `mcpClient` para generar un complemento Genkit para un servidor MCP. Por ejemplo, para usar [el servidor de sistema de archivos de](https://github.com/modelcontextprotocol/servers/tree/main/src/filesystem) ejemplo de MCP:\n\n```ts\nimport { genkit } from 'genkit';\nimport { mcpClient } from 'genkitx-mcp';\n\n// the filesystem server requires one or more allowed directories\nconst ALLOWED_DIRS = ['/Users/yourusername/Desktop'];\n\nconst filesystemClient = mcpClient({\n name: 'filesystem',\n serverProcess: {\n command: 'npx',\n args: ['-y', '@modelcontextprotocol/server-everything', ...ALLOWED_DIRS],\n },\n});\n\nconst ai = genkit({\n plugins: [\n filesystemClient /* ... other plugins such as model providers ...*/,\n ],\n});\n```\n\nLa mayoría de los servidores MCP están diseñados para ejecutarse como procesos generados en la misma máquina mediante el transporte `stdio` . Al proporcionar la opción `serverProcess` , se especifican el comando, los argumentos y las variables de entorno para generar el servidor como subproceso.\n\n### Opciones de mcpClient()\n\n* **`name`** : (obligatorio) El nombre de este cliente, que asigna espacios de nombres a sus herramientas y avisos.\n* **`version`** : (opcional) Número de versión del cliente. El valor predeterminado es \"1.0.0\".\n* Debe proporcionar uno de los siguientes:\n * **`serverProcess`** : Parámetros para iniciar un proceso de servidor local utilizando el transporte MCP stdio.\n * **`command`** : Ruta de comandos del shell para iniciar el servidor MCP. Puede ser, por ejemplo, `npx` o `uvx` para descargar y ejecutar el servidor desde un gestor de paquetes.\n * **`args`** : (opcional) Matriz de argumentos de cadena para pasar al comando.\n * **`env`** : (opcional) Objeto de valor clave de las variables de entorno que se pasarán al comando.\n * **`serverUrl`** : la URL de un servidor remoto al que conectarse mediante el transporte SSE MCP.\n * \\*\\* `serverWebsocketUrl` : la URL de un servidor remoto al que conectarse mediante el transporte MCP de WebSocket.\n * **`transport`** : un objeto de transporte MCP existente para conectarse al servidor.\n* **`rawToolResponses`** : (opcional) Indicador booleano. Si es `true` , las respuestas de la herramienta se devuelven en su formato MCP original; de lo contrario, se procesan para compatibilidad con Genkit.\n\n### Uso de acciones MCP\n\nEl cliente MCP de Genkit detecta automáticamente las herramientas y los avisos disponibles y los registra en Genkit, haciéndolos disponibles dondequiera que se puedan usar otras herramientas y avisos. Para acceder a los recursos, se registran las herramientas especiales `list_resources` y `read_resource` , que accederán a los recursos del servidor.\n\nTodas las acciones de MCP tienen un espacio de nombres bajo el nombre que usted proporciona, por lo que un cliente llamado `filesystem` registrará herramientas como `filesystem/read_file` .\n\n### Respuestas de la herramienta\n\nLas herramientas MCP devuelven una matriz `content` , a diferencia de una respuesta estructurada como la mayoría de las herramientas Genkit. El complemento MCP de Genkit intenta analizar y convertir el contenido devuelto:\n\n1. Si el contenido es texto y JSON válido, se analiza y se devuelve el JSON.\n2. Si el contenido es texto y no JSON válido, se devuelve el texto.\n3. Si el contenido tiene una sola parte que no es texto, se devuelve.\n4. Si el contenido tiene partes múltiples o mixtas, se devuelve la respuesta del contenido completo.\n\n## Servidor MCP\n\nTambién puedes exponer todas las herramientas y avisos de una instancia de Genkit como un servidor MCP:\n\n```ts\nimport { genkit, z } from 'genkit';\nimport { mcpServer } from 'genkitx-mcp';\n\nconst ai = genkit({});\n\nai.defineTool(\n {\n name: 'add',\n description: 'add two numbers together',\n inputSchema: z.object({ a: z.number(), b: z.number() }),\n outputSchema: z.number(),\n },\n async ({ a, b }) => {\n return a + b;\n }\n);\n\nai.definePrompt(\n {\n name: \"happy\",\n description: \"everybody together now\",\n input: {\n schema: z.object({\n action: z.string().default(\"clap your hands\").optional(),\n }),\n },\n },\n `If you're happy and you know it, {{action}}.`\n);\n\nmcpServer(ai, { name: 'example_server', version: '0.0.1' }).start();\n```\n\nLo anterior iniciará un servidor MCP con el transporte stdio que expone una herramienta llamada `add` y un indicador llamado `happy` . Para iniciar el servidor con un transporte diferente, use `mcpServer(...).start(otherTransport)` .\n\n### Limitaciones conocidas\n\n* Las solicitudes de MCP solo pueden tomar parámetros de cadena, por lo que las entradas a los esquemas deben ser objetos con solo valores de propiedad de cadena.\n* Los mensajes de MCP solo admiten mensajes `user` y `model` . Los mensajes `system` no son compatibles.\n* Los mensajes de MCP solo admiten un único \"tipo\" dentro de un mensaje, por lo que no es posible mezclar medios y texto en el mismo mensaje.\n\n### Probando su servidor MCP\n\nPuedes probar tu servidor MCP con el inspector oficial. Por ejemplo, si el código de tu servidor se compila en `dist/index.js` , podrías ejecutar:\n\n```\nnpx @modelcontextprotocol/inspector dist/index.js\n```\n\nUna vez que inicie el inspector, podrá enumerar indicaciones y acciones y probarlas manualmente.","Spanish","es-ES",{"_27":51,"_31":55,"_36":56},{"_29":52,"_31":53,"_33":54},"[Genkit](https://github.com/firebase/genkit)と Model Context Protocol (MCP) 間の統合を提供します。","ジェンキットMCP","# ジェンキットMCP\n\n> \\[！警告]\\\n> このプラグインは実験的なものであり、長期的にはサポートされない可能性があり、API は頻繁に重大な変更を受ける可能性があります。\n\nこのプラグインは、Genkitと[モデルコンテキストプロトコル](https://modelcontextprotocol.io)（MCP）の統合を提供します。MCPは、開発者がクライアントにツール、リソース、プロンプトを提供する「サーバー」を構築するためのオープンスタンダードです。Genkit MCPを使用すると、Genkit開発者はクライアントとしてMCPのツール、プロンプト、リソースを利用するだけでなく、サーバーとしてツールやプロンプトを提供することもできます。\n\n## インストール\n\n始めるには、Genkit と MCP プラグインが必要です。\n\n```bash\nnpm i genkit genkitx-mcp\n```\n\n## MCPクライアント\n\nMCPクライアントを作成するには、 `mcpClient`関数を呼び出して、MCPサーバー用のGenkitプラグインを生成します。例えば、MCPのサン��ル[ファイルシステムサーバー](https://github.com/modelcontextprotocol/servers/tree/main/src/filesystem)を使用する場合は、次のようにします。\n\n```ts\nimport { genkit } from 'genkit';\nimport { mcpClient } from 'genkitx-mcp';\n\n// the filesystem server requires one or more allowed directories\nconst ALLOWED_DIRS = ['/Users/yourusername/Desktop'];\n\nconst filesystemClient = mcpClient({\n name: 'filesystem',\n serverProcess: {\n command: 'npx',\n args: ['-y', '@modelcontextprotocol/server-everything', ...ALLOWED_DIRS],\n },\n});\n\nconst ai = genkit({\n plugins: [\n filesystemClient /* ... other plugins such as model providers ...*/,\n ],\n});\n```\n\nほとんどのMCPサーバーは、 `stdio`トランスポートを使用して、同一マシン上で生成されたプロセスとして実行されるように構築されています。serverProcessオプションを指定すると、サーバー`serverProcess`サブプロセスとして生成するためのコマンド、引数、および環境変数が指定されます。\n\n### mcpClient() オプション\n\n* **`name`** : (必須) このクライアントの名前。ツールとプロンプトの名前空間を指定します。\n* **`version`** : (オプション) クライアントのバージョン番号。デフォルトは「1.0.0」です。\n* 次のいずれかを指定する必要があります:\n * **`serverProcess`** : stdio MCP トランスポートを使用してローカルサーバープロセスを起動するためのパラメーター。\n * **`command`** : MCPサーバーを起動するためのシェルコマンドパス。パッケージマネージャーからサーバーをダウンロードして実行する場合は、例えば`npx`や`uvx`などを使用します。\n * **`args`** : (オプション) コマンドに渡す文字列引数の配列。\n * **`env`** : (オプション) コマンドに渡す環境変数のキー値オブジェクト。\n * **`serverUrl`** : SSE MCP トランスポートを使用して接続するリモートサーバーの URL。\n * \\*\\* `serverWebsocketUrl` : WebSocket MCP トランスポートを使用して接続するリモートサーバーの URL。\n * **`transport`** : サーバーに接続するための既存の MCP トランスポートオブジェクト。\n* **`rawToolResponses`** : (オプション) ブール型フラグ。true `true`場合、ツールレスポンスは生のMCP形式で返されます。それ以外の場合は、Genkitとの互換性のために処理されます。\n\n### MCPアクションの使用\n\nGenkit MCPクライアントは、利用可能なツールとプロンプトを自動的に検出し、Genkitに登録します。これにより、他のツールやプロンプトが使用できる場所であればどこでも利用できるようになります。リソースにアクセスするために、サーバーのリソースにアクセスするための特別な`list_resources`と`read_resource`ツールが登録されています。\n\nすべての MCP アクションは指定した名前で名前空間化されるため、 `filesystem`というクライアントは`filesystem/read_file`などのツールを登録します。\n\n### ツールの応答\n\nMCPツールは、多くのGenkitツールのような構造化されたレスポンスではなく、 `content`配列を返します。Genkit MCPプラグインは、返されたコンテンツを解析して強制的に変換しようとします。\n\n1. コンテンツがテキストで有効な JSON の場合、JSON が解析されて返されます。\n2. コンテンツがテキストであり、有効な JSON ではない場合は、テキストが返されます。\n3. コンテンツにテキスト以外の部分が 1 つだけある場合は、それが返されます。\n4. コンテンツに複数の部分または混合部分がある場合は、完全なコンテンツ応答が返されます。\n\n## MCPサーバー\n\nGenkit インスタンスからのすべてのツールとプロンプトを MCP サーバーとして公開することもできます。\n\n```ts\nimport { genkit, z } from 'genkit';\nimport { mcpServer } from 'genkitx-mcp';\n\nconst ai = genkit({});\n\nai.defineTool(\n {\n name: 'add',\n description: 'add two numbers together',\n inputSchema: z.object({ a: z.number(), b: z.number() }),\n outputSchema: z.number(),\n },\n async ({ a, b }) => {\n return a + b;\n }\n);\n\nai.definePrompt(\n {\n name: \"happy\",\n description: \"everybody together now\",\n input: {\n schema: z.object({\n action: z.string().default(\"clap your hands\").optional(),\n }),\n },\n },\n `If you're happy and you know it, {{action}}.`\n);\n\nmcpServer(ai, { name: 'example_server', version: '0.0.1' }).start();\n```\n\n上記のコマンドは、stdioトランスポートを使用してMCPサーバーを起動します。stdioトランスポートは、 `add`というツールと`happy`というプロンプトを表示します。別のトランスポートを使用してサーバーを起動するには、 `mcpServer(...).start(otherTransport)`使用します。\n\n### 既知の制限事項\n\n* MCP プロンプトは文字列パラメータのみを受け入れることができるため、スキーマへの入力は文字列プロパティ値のみを持つオブジェクトである必要があります。\n* MCP プロンプトは`user`と`model`メッセージのみをサポートします。 `system`メッセージはサポートされません。\n* MCP プロンプトはメッセージ内で単一の「タイプ」のみをサポートするため、同じメッセージ内でメディアとテキストを混在させることはできません。\n\n### MCPサーバーのテスト\n\n公式インスペクタを使ってMCPサーバーをテストできます。例えば、サーバーコードが`dist/index.js`にコンパイルされている場合は、次のコマンドを実行できます。\n\n```\nnpx @modelcontextprotocol/inspector dist/index.js\n```\n\nインスペクターを起動すると、プロンプトとアクションを一覧表示し、手動でテストすることができます。","Japanese","ja-JP",{"_27":58,"_31":62,"_36":63},{"_29":59,"_31":60,"_33":61},"[Genkit](https://github.com/firebase/genkit) 과 Model Context Protocol(MCP) 간의 통합을 제공합니다.","젠킷 MCP","# 젠킷 MCP\n\n> \\[!경고]\\\n> 이 플러그인은 실험적이므로 장기적으로 지원되지 않을 수 있으며 API는 자주 중단되는 변경 사항을 받을 수 있습니다.\n\n이 플러그인은 Genkit과 [모델 컨텍스트 프로토콜](https://modelcontextprotocol.io) (MCP) 간의 통합을 제공합니다. MCP는 개발자가 클라이언트에 도구, 리소스 및 프롬프트를 제공하는 \"서버\"를 구축할 수 있도록 하는 개방형 표준입니다. Genkit MCP를 통해 Genkit 개발자는 MCP 도구, 프롬프트 및 리소스를 클라이언트로 사용하는 동시에 도구와 프롬프트를 서버로 제공할 수 있습니다.\n\n## 설치\n\n시작하려면 Genkit과 MCP 플러그인이 필요합니다.\n\n지엑스피1\n\n## MCP 클라이언트\n\nMCP 클라이언트를 생성하려면 `mcpClient` 함수를 호출하여 MCP 서버용 Genkit 플러그인을 생성합니다. 예를 들어, MCP의 [파일 시스템 서버](https://github.com/modelcontextprotocol/servers/tree/main/src/filesystem) 예제를 사용하려면 다음과 같이 합니다.\n\n```ts\nimport { genkit } from 'genkit';\nimport { mcpClient } from 'genkitx-mcp';\n\n// the filesystem server requires one or more allowed directories\nconst ALLOWED_DIRS = ['/Users/yourusername/Desktop'];\n\nconst filesystemClient = mcpClient({\n name: 'filesystem',\n serverProcess: {\n command: 'npx',\n args: ['-y', '@modelcontextprotocol/server-everything', ...ALLOWED_DIRS],\n },\n});\n\nconst ai = genkit({\n plugins: [\n filesystemClient /* ... other plugins such as model providers ...*/,\n ],\n});\n```\n\n대부분의 MCP 서버는 `stdio` 전송을 사용하여 동일한 머신에서 생성된 프로세스로 실행되도록 설계되었습니다. `serverProcess` 옵션을 제공하면 서버를 하위 프로세스로 생성하는 데 필요한 명령, 인수 및 환경 변수가 지정됩니다.\n\n### mcpClient() 옵션\n\n* **`name`** : (필수) 이 클라이언트의 이름으로, 도구와 프롬프트의 네임스페이스입니다.\n* **`version`** : (선택 사항) 클라이언트 버전 번호입니다. 기본값은 \"1.0.0\"입니다.\n* 다음 중 하나를 제공해야 합니다.\n * **`serverProcess`** : stdio MCP 전송을 사용하여 로컬 서버 프로세스를 시작하기 위한 매개변수입니다.\n * **`command`** : MCP 서버를 시작하기 위한 셸 명령 경로입니다. 예를 들어 `npx` 또는 `uvx` 사용하여 패키지 관리자에서 서버를 다운로드하고 실행할 수 있습니다.\n * **`args`** : (선택 사항) 명령에 전달할 문자열 인수 배열입니다.\n * **`env`** : (선택 사항) 명령에 전달할 환경 변수의 키 값 객체입니다.\n * **`serverUrl`** : SSE MCP 전송을 사용하여 연결할 원격 서버의 URL입니다.\n * \\*\\* `serverWebsocketUrl` : WebSocket MCP 전송을 사용하여 연결할 원격 서버의 URL입니다.\n * **`transport`** : 서버에 연결하기 위한 기존 MCP 전송 개체입니다.\n* **`rawToolResponses`** : (선택 사항) 부울 플래그입니다. `true` 이면 도구 응답이 원시 MCP 형식으로 반환되고, 그렇지 않으면 Genkit 호환성을 위해 처리됩니다.\n\n### MCP 작업 사용\n\nGenkit MCP 클라이언트는 사용 가능한 도구와 프롬프트를 자동으로 검색하고 Genkit에 등록하여 다른 도구와 프롬프트를 사용할 수 있는 모든 곳에서 사용할 수 있도록 합니다. 리소스에 액세스하려면 서버의 리소스에 액세스하는 특수 도구인 `list_resources` 및 `read_resource` 가 등록됩니다.\n\n모든 MCP 작업은 사용자가 제공한 이름으로 네임스페이스가 지정되므로 `filesystem` 이라는 클라이언트는 `filesystem/read_file` 과 같은 도구를 등록합니다.\n\n### 도구 응답\n\nMCP 도구는 대부분의 Genkit 도구처럼 구조화된 응답이 아닌 `content` 배열을 반환합니다. Genkit MCP 플러그인은 반환된 콘텐츠를 파싱하고 강제 변환하려고 시도합니다.\n\n1. 콘텐츠가 텍스트이고 유효한 JSON인 경우 JSON이 구문 분석되어 반환됩니다.\n2. 콘텐츠가 텍스트이고 유효한 JSON이 아닌 경우 텍스트가 반환됩니다.\n3. 콘텐츠에 텍스트가 아닌 부분이 하나만 있는 경우 해당 부분이 반환됩니다.\n4. 콘텐츠에 여러 부분/혼합 부분이 있는 경우 전체 콘텐츠 응답이 반환됩니다.\n\n## MCP 서버\n\nGenkit 인스턴스의 모든 도구와 프롬프트를 MCP 서버로 공개할 수도 있습니다.\n\n```ts\nimport { genkit, z } from 'genkit';\nimport { mcpServer } from 'genkitx-mcp';\n\nconst ai = genkit({});\n\nai.defineTool(\n {\n name: 'add',\n description: 'add two numbers together',\n inputSchema: z.object({ a: z.number(), b: z.number() }),\n outputSchema: z.number(),\n },\n async ({ a, b }) => {\n return a + b;\n }\n);\n\nai.definePrompt(\n {\n name: \"happy\",\n description: \"everybody together now\",\n input: {\n schema: z.object({\n action: z.string().default(\"clap your hands\").optional(),\n }),\n },\n },\n `If you're happy and you know it, {{action}}.`\n);\n\nmcpServer(ai, { name: 'example_server', version: '0.0.1' }).start();\n```\n\n위 코드는 `add` 라는 도구와 `happy` 라는 프롬프트를 노출하는 stdio 전송을 사용하여 MCP 서버를 시작합니다. 다른 전송으로 서버를 시작하려면 `mcpServer(...).start(otherTransport)` 사용합니다.\n\n### 알려진 제한 사항\n\n* MCP 프롬프트는 문자열 매개변수만 사용할 수 있으므로 스키마에 대한 입력은 문자열 속성 값만 있는 개체여야 합니다.\n* MCP 프롬프트는 `user` 및 `model` 메시지만 지원합니다. `system` 메시지는 지원되지 않습니다.\n* MCP 프롬프트는 메시지 내에서 단일 \"유형\"만 지원하므로 동일한 메시지에 미디어와 텍스트를 혼합할 수 없습니다.\n\n### MCP 서버 테스트\n\n공식 검사기를 사용하여 MCP 서버를 테스트할 수 있습니다. 예를 들어, 서버 코드가 `dist/index.js` 파일로 컴파일된 경우 다음을 실행할 수 있습니다.\n\n```\nnpx @modelcontextprotocol/inspector dist/index.js\n```\n\n검사기를 시작하면 프롬프트와 작업을 나열하고 수동으로 테스트할 수 있습니다.","Korean","ko-KR",{"_27":65,"_31":69,"_36":70},{"_29":66,"_31":67,"_33":68},"Обеспечивает интеграцию между [Genkit](https://github.com/firebase/genkit) и протоколом контекста модели (MCP).","Генкит MCP","# Генкит MCP\n\n> \\[!ПРЕДУПРЕЖДЕНИЕ]\\\n> Этот плагин является экспериментальным, то есть он не будет поддерживаться в долгосрочной перспективе, а API-интерфейсы подвержены более частым критическим изменениям.\n\nЭтот плагин обеспечивает интеграцию между Genkit и [Model Context Protocol](https://modelcontextprotocol.io) (MCP). MCP — это открытый стандарт, позволяющий разработчикам создавать «серверы», которые предоставляют инструменты, ресурсы и подсказки клиентам. Genkit MCP позволяет разработчикам Genkit как использовать инструменты, подсказки и ресурсы MCP в качестве клиента, так и предоставлять инструменты и подсказки в качестве сервера.\n\n## Установка\n\nДля начала вам понадобится Genkit и плагин MCP:\n\n```bash\nnpm i genkit genkitx-mcp\n```\n\n## Клиент МСР\n\nЧтобы создать MCP-клиент, вы вызываете функцию `mcpClient` для генерации плагина Genkit для MCP-сервера. Например, чтобы использовать пример [сервера файловой системы](https://github.com/modelcontextprotocol/servers/tree/main/src/filesystem) MCP:\n\n```ts\nimport { genkit } from 'genkit';\nimport { mcpClient } from 'genkitx-mcp';\n\n// the filesystem server requires one or more allowed directories\nconst ALLOWED_DIRS = ['/Users/yourusername/Desktop'];\n\nconst filesystemClient = mcpClient({\n name: 'filesystem',\n serverProcess: {\n command: 'npx',\n args: ['-y', '@modelcontextprotocol/server-everything', ...ALLOWED_DIRS],\n },\n});\n\nconst ai = genkit({\n plugins: [\n filesystemClient /* ... other plugins such as model providers ...*/,\n ],\n});\n```\n\nБольшинство серверов MCP созданы для запуска в качестве порожденных процессов на той же машине с использованием транспорта `stdio` . Когда вы указываете опцию `serverProcess` , вы указываете команду, аргументы и переменные среды для порождения сервера как подпроцесса.\n\n### Параметры mcpClient()\n\n* **`name`** : (обязательно) Имя этого клиента, которое определяет пространство имен его инструментов и подсказок.\n* **`version`** : (необязательно) Номер версии клиента. По умолчанию \"1.0.0\".\n* Вам необходимо предоставить один из следующих документов:\n * **`serverProcess`** : Параметры для запуска локального серверного процесса с использованием транспорта stdio MCP.\n * **`command`** : Путь к команде оболочки для запуска сервера MCP. Может быть, например, `npx` или `uvx` для загрузки и запуска сервера из менеджера пакетов.\n * **`args`** : (необязательно) Массив строковых аргументов для передачи команде.\n * **`env`** : (необязательно) объект значения ключа переменных среды для передачи команде.\n * **`serverUrl`** : URL-адрес удаленного сервера для подключения с использованием транспорта SSE MCP.\n * \\*\\* `serverWebsocketUrl` : URL-адрес удаленного сервера для подключения с использованием транспорта WebSocket MCP.\n * **`transport`** : существующий объект транспорта MCP для подключения к серверу.\n* **`rawToolResponses`** : (необязательно) Логический флаг. Если `true` , ответы инструмента возвращаются в необработанном формате MCP; в противном случае они обрабатываются для совместимости с Genkit.\n\n### Использование действий MCP\n\nКлиент Genkit MCP автоматически обнаруживает доступные инструменты и подсказки и регистрирует их в Genkit, делая их доступными везде, где могут использоваться другие инструменты и подсказки. Для доступа к ресурсам регистрируются специальные инструменты `list_resources` и `read_resource` , которые будут получать доступ к ресурсам сервера.\n\nВсе действия MCP размещаются в пространстве имен под указанным вами именем, поэтому клиент с именем `filesystem` будет регистрировать такие инструменты, как `filesystem/read_file` .\n\n### Ответы инструментов\n\nИнструменты MCP возвращают массив `content` , а не структурированный ответ, как большинство инструментов Genkit. Плагин Genkit MCP пытается проанализировать и привести возвращенный контент к следующему виду:\n\n1. Если содержимое представляет собой текст и допустимый JSON, JSON анализируется и возвращается.\n2. Если содержимое представляет собой текст, а не допустимый JSON, возвращается текст.\n3. Если содержимое содержит одну нетекстовую часть, она возвращается.\n4. Если содержимое состоит из нескольких/смешанных частей, возвращается полный ответ содержимого.\n\n## MCP-сервер\n\nВы также можете предоставить доступ ко всем инструментам и подсказкам из экземпляра Genkit как к серверу MCP:\n\n```ts\nimport { genkit, z } from 'genkit';\nimport { mcpServer } from 'genkitx-mcp';\n\nconst ai = genkit({});\n\nai.defineTool(\n {\n name: 'add',\n description: 'add two numbers together',\n inputSchema: z.object({ a: z.number(), b: z.number() }),\n outputSchema: z.number(),\n },\n async ({ a, b }) => {\n return a + b;\n }\n);\n\nai.definePrompt(\n {\n name: \"happy\",\n description: \"everybody together now\",\n input: {\n schema: z.object({\n action: z.string().default(\"clap your hands\").optional(),\n }),\n },\n },\n `If you're happy and you know it, {{action}}.`\n);\n\nmcpServer(ai, { name: 'example_server', version: '0.0.1' }).start();\n```\n\nВышеприведенный код запустит сервер MCP с транспортом stdio, который предоставляет инструмент `add` и приглашение `happy` . Чтобы запустить сервер с другим транспортом, используйте `mcpServer(...).start(otherTransport)` .\n\n### Известные ограничения\n\n* Приглашения MCP могут принимать только строковые параметры, поэтому входные данные для схем должны быть объектами, имеющими только строковые значения свойств.\n* Подсказки MCP поддерживают только сообщения `user` и `model` . `system` сообщения не поддерживаются.\n* Подсказки MCP поддерживают только один «тип» в сообщении, поэтому вы не можете смешивать медиа и текст в одном сообщении.\n\n### Тестирование вашего MCP-сервера\n\nВы можете протестировать свой сервер MCP с помощью официального инспектора. Например, если ваш код сервера скомпилирован в `dist/index.js` , вы можете запустить:\n\n```\nnpx @modelcontextprotocol/inspector dist/index.js\n```\n\nПосле запуска инспектора вы можете составить список подсказок и действий и протестировать их вручную.","Russian","ru-RU",{"_27":72,"_31":75,"_36":76},{"_29":73,"_31":32,"_33":74},"提供[Genkit](https://github.com/firebase/genkit)和模型上下文协议 (MCP) 之间的集成。","# Genkit MCP\n\n> \\[！警告]\\\n> 该插件是实验性的，这意味着它可能不会得到长期支持，并且 API 可能会经常发生重大变化。\n\n此插件提供 Genkit 与[模型上下文协议](https://modelcontextprotocol.io)(MCP) 之间的集成。MCP 是一个开放标准，允许开发者构建“服务器”，向客户端提供工具、资源和提示。Genkit MCP 允许 Genkit 开发者既可以作为客户端使用 MCP 工具、提示和资源，也可以作为服务器提供工具和提示。\n\n## 安装\n\n首先，您需要 Genkit 和 MCP 插件：\n\n```bash\nnpm i genkit genkitx-mcp\n```\n\n## MCP 客户端\n\n要创建 MCP 客户端，请调用`mcpClient`函数为 MCP 服务器生成 Genkit 插件。例如，要使用 MCP 的示例[文件系统服务器](https://github.com/modelcontextprotocol/servers/tree/main/src/filesystem)：\n\n```ts\nimport { genkit } from 'genkit';\nimport { mcpClient } from 'genkitx-mcp';\n\n// the filesystem server requires one or more allowed directories\nconst ALLOWED_DIRS = ['/Users/yourusername/Desktop'];\n\nconst filesystemClient = mcpClient({\n name: 'filesystem',\n serverProcess: {\n command: 'npx',\n args: ['-y', '@modelcontextprotocol/server-everything', ...ALLOWED_DIRS],\n },\n});\n\nconst ai = genkit({\n plugins: [\n filesystemClient /* ... other plugins such as model providers ...*/,\n ],\n});\n```\n\n大多数 MCP 服务器都构建为在同一台机器上使用`stdio`传输方式作为子进程运行。当您提供`serverProcess`选项时，您将指定将服务器作为子进程生成的命令、参数和环境变量。\n\n### mcpClient() 选项\n\n* **`name`** ：（必需）此客户端的名称，其工具和提示的命名空间。\n* **`version`** ：（可选）客户端版本号。默认为“1.0.0”。\n* 您必须提供以下之一：\n * **`serverProcess`** ：使用 stdio MCP 传输启动本地服务器进程的参数。\n * **`command`** ：用于启动 MCP 服务器的 Shell 命令路径。例如，可以使用`npx`或`uvx`命令从包管理器下载并运行服务器。\n * **`args`** ：（可选）传递给命令的字符串参数数组。\n * **`env`** ：（可选）传递给命令的环境变量的键值对象。\n * **`serverUrl`** ：使用 SSE MCP 传输连接的远程服务器的 URL。\n * \\*\\* `serverWebsocketUrl` ：使用 WebSocket MCP 传输连接的远程服务器的 URL。\n * **`transport`** ：用于连接服务器的现有 MCP 传输对象。\n* **`rawToolResponses`** ：（可选）布尔值标志。如果为`true` ，则工具响应将以原始 MCP 格式返回；否则，将对其进行处理以兼容 Genkit。\n\n### 使用 MCP 操作\n\nGenkit MCP 客户端会自动发现可用的工具和提示符，并将其注册到 Genkit，以便在其他工具和提示符可以使用的任何地方使用它们。为了访问资源，需要注册特殊的`list_resources`和`read_resource`工具来访问服务器的资源。\n\n所有 MCP 操作都以您提供的名称命名，因此名为`filesystem`的客户端将注册诸如`filesystem/read_file`之类的工具。\n\n### 工具响应\n\nMCP 工具返回的是`content`数组，而不是像大多数 Genkit 工具那样返回结构化的响应。Genkit MCP 插件会尝试解析并强制执行返回的内容：\n\n1. 如果内容是文本和有效的 JSON，则解析并返回 JSON。\n2. 如果内容是文本并且不是有效的 JSON，则返回文本。\n3. 如果内容有一个非文本部分，则返回该部分。\n4. 如果内容有多个/混合部分，则返回完整的内容响应。\n\n## MCP 服务器\n\n您还可以将 Genkit 实例中的所有工具和提示公开为 MCP 服务器：\n\n```ts\nimport { genkit, z } from 'genkit';\nimport { mcpServer } from 'genkitx-mcp';\n\nconst ai = genkit({});\n\nai.defineTool(\n {\n name: 'add',\n description: 'add two numbers together',\n inputSchema: z.object({ a: z.number(), b: z.number() }),\n outputSchema: z.number(),\n },\n async ({ a, b }) => {\n return a + b;\n }\n);\n\nai.definePrompt(\n {\n name: \"happy\",\n description: \"everybody together now\",\n input: {\n schema: z.object({\n action: z.string().default(\"clap your hands\").optional(),\n }),\n },\n },\n `If you're happy and you know it, {{action}}.`\n);\n\nmcpServer(ai, { name: 'example_server', version: '0.0.1' }).start();\n```\n\n上述代码将启动一个带有 stdio 传输的 MCP 服务器，该服务器会暴露一个名为`add`的工具和一个名为`happy`的提示符。要使用其他传输方式启动服务器，请使用`mcpServer(...).start(otherTransport)` 。\n\n### 已知限制\n\n* MCP 提示只能采用字符串参数，因此模式的输入必须是仅具有字符串属性值的对象。\n* MCP 提示仅支持`user`和`model`消息。不支持`system`消息。\n* MCP 提示仅支持消息中的单一“类型”，因此您不能在同一消息中混合媒体和文本。\n\n### 测试您的 MCP 服务器\n\n你可以使用官方检查器测试你的 MCP 服务器。例如，如果你的服务器代码编译为`dist/index.js` ，你可以运行：\n\n```\nnpx @modelcontextprotocol/inspector dist/index.js\n```\n\n一旦启动检查器，您就可以列出提示和操作并手动测试它们。","Chinese","zh-CN","githubUser","locale",{"_80":81,"_13":82,"_91":40,"_92":93,"_16":94,"_95":96,"_31":32,"_108":109,"_116":-5,"_117":118,"_119":120,"_171":172,"_177":-5,"_103":152,"_178":156,"_106":179,"_180":181},"addedAt","2025-01-20T17:10:37.200606Z",[83,84,85,86,87,88,89,90],"author:official","category:developer-tools","category:app-automation","hosting:hybrid","language:typescript","security_grade:a","quality_grade:a","license_grade:a","descriptionMarkdown","descriptionPlainText","Provides integration between Genkit and the Model Context Protocol (MCP).",694,"integrations",[97],{"_98":99,"_29":105,"_106":107},"brand",{"_100":101,"_31":102,"_103":104},"logoSvgPath","M12.367 2.453a.822.822 0 0 0-.576.238L.241 14.213a.822.822 0 0 0-.241.584v.066c0-.323.209-.608.516-.709l7.275-2.318a2.437 2.437 0 0 0 1.584-1.592l2.318-7.267a.757.757 0 0 1 .719-.524zM0 14.863v5.047c0 .904.733 1.637 1.637 1.637h20.726c.904 0 1.637-.733 1.637-1.637V4.09c0-.904-.733-1.637-1.637-1.637h-9.951v.5l.088 9.861c.01 1.175-.962 2.14-2.137 2.14L0 14.862zM12 3.66l-2.148 6.735v.001a2.94 2.94 0 0 1-1.909 1.916l-6.716 2.141h9.136c.905 0 1.638-.734 1.637-1.639zm-10.363.975c-.905 0-1.638.734-1.637 1.638v7.473l9.135-9.111Z","Files","slug","files","Allows integration with the filesystem server through MCP, enabling reading files from allowed directories","uid","b3bssdahbo","namespace",{"_110":111,"_103":115},"owner",{"_112":113},"githubAccount",{"_114":115},"login","firebase","readmeMarkdown","redditPostExternalId","1i5wvj3","repository",{"_121":122,"_136":137,"_141":142,"_162":-5,"_163":164,"_165":166,"_170":-5},"argumentsJsonSchema",{"_123":124,"_133":134,"_128":135},"properties",{"_125":126,"_130":131},"PUBMED_EMAIL",{"_29":127,"_128":129},"Your email address (required by NCBI)","type","string","PUBMED_API_KEY",{"_29":132,"_128":129},"Optional API key for higher rate limits","required",[125],"object","githubProject",{"_33":138,"_139":140},"# Genkit MCP\n\nSee [Genkit MCP documentation](https://genkit.dev/docs/model-context-protocol/).\n\nThis plugin provides integration between Genkit and the [Model Context Protocol](https://modelcontextprotocol.io) (MCP). MCP is an open standard allowing developers to build \"servers\" which provide tools, resources, and prompts to clients. Genkit MCP allows Genkit developers to:\n- Consume MCP tools, prompts, and resources as a client using `createMcpHost` or `createMcpClient`.\n- Provide Genkit tools and prompts as an MCP server using `createMcpServer`.\n\n## Installation\n\nTo get started, you'll need Genkit and the MCP plugin:\n\n```bash\nnpm i genkit @genkit-ai/mcp\n```\n\n## MCP Host\n\nTo connect to one or more MCP servers, you use the `createMcpHost` function. This function returns a `GenkitMcpHost` instance that manages connections to the configured MCP servers.\n\n```ts\nimport { googleAI } from '@genkit-ai/google-genai';\nimport { createMcpHost } from '@genkit-ai/mcp';\nimport { genkit } from 'genkit';\n\nconst mcpHost = createMcpHost({\n name: 'myMcpClients', // A name for the host plugin itself\n mcpServers: {\n // Each key (e.g., 'fs', 'git') becomes a namespace for the server's tools.\n fs: {\n command: 'npx',\n args: ['-y', '@modelcontextprotocol/server-filesystem', process.cwd()],\n },\n memory: {\n command: 'npx',\n args: ['-y', '@modelcontextprotocol/server-memory'],\n },\n },\n});\n\nconst ai = genkit({\n plugins: [googleAI()],\n});\n\n(async () => {\n // Provide MCP tools to the model of your choice.\n const { text } = await ai.generate({\n model: googleAI.model('gemini-2.0-flash'),\n prompt: `Analyze all files in ${process.cwd()}.`,\n tools: await mcpHost.getActiveTools(ai),\n resources: await mcpHost.getActiveResources(ai),\n });\n\n console.log(text);\n\n await mcpHost.close();\n})();\n```\n\nThe `createMcpHost` function initializes a `GenkitMcpHost` instance, which handles the lifecycle and communication with the defined MCP servers.\n\n### `createMcpHost()` Options\n\n- **`name`**: (optional, string) A name for the MCP host plugin itself. Defaults to 'genkitx-mcp'.\n- **`version`**: (optional, string) The version of the MCP host plugin. Defaults to \"1.0.0\".\n- **`rawToolResponses`**: (optional, boolean) When `true`, tool responses are returned in their raw MCP format; otherwise, they are processed for Genkit compatibility. Defaults to `false`.\n- **`mcpServers`**: (required, object) An object where each key is a client-side name (namespace) for an MCP server, and the value is the configuration for that server.\n\n Each server configuration object can include:\n - **`disabled`**: (optional, boolean) If `true`, this server connection will not be attempted. Defaults to `false`.\n - One of the following server connection configurations:\n - Parameters for launching a local server process using the stdio MCP transport.\n - **`command`**: (required, string) Shell command path for launching the MCP server (e.g., `npx`, `python`).\n - **`args`**: (optional, string[]) Array of string arguments to pass to the command.\n - **`env`**: (optional, Record) Key-value object of environment variables.\n - **`url`**: (string) The URL of a remote server to connect to using the Streamable HTTP MCP transport.\n - **`transport`**: An existing MCP transport object for connecting to the server.\n\n\n## MCP Client (Single Server)\n\nFor scenarios where you only need to connect to a single MCP server, or prefer to manage client instances individually, you can use `createMcpClient`.\n\n```ts\nimport { googleAI } from '@genkit-ai/google-genai';\nimport { createMcpClient } from '@genkit-ai/mcp';\nimport { genkit } from 'genkit';\n\nconst myFsClient = createMcpClient({\n name: 'myFileSystemClient', // A unique name for this client instance\n mcpServer: {\n command: 'npx',\n args: ['-y', '@modelcontextprotocol/server-filesystem', process.cwd()],\n },\n // rawToolResponses: true, // Optional: get raw MCP responses\n});\n\n// In your Genkit configuration:\nconst ai = genkit({\n plugins: [googleAI()],\n});\n\n(async () => {\n await myFsClient.ready();\n\n // Retrieve tools from this specific client\n const fsTools = await myFsClient.getActiveTools(ai);\n\n const { text } = await ai.generate({\n model: googleAI.model('gemini-2.0-flash'), // Replace with your model\n prompt: 'List files in ' + process.cwd(),\n tools: fsTools,\n });\n console.log(text);\n\n await myFsClient.disable();\n})();\n```\n\n### `createMcpClient()` Options\n\nThe `createMcpClient` function takes an `McpClientOptions` object:\n- **`name`**: (required, string) A unique name for this client instance. This name will be used as the namespace for its tools and prompts.\n- **`version`**: (optional, string) Version for this client instance. Defaults to \"1.0.0\".\n- Additionally, it supports all options from `McpServerConfig` (e.g., `disabled`, `rawToolResponses`, and transport configurations), as detailed in the `createMcpHost` options section.\n\n### Using MCP Actions (Tools, Prompts)\n\nBoth `GenkitMcpHost` (via `getActiveTools()`) and `GenkitMcpClient` (via `getActiveTools()`) discover available tools from their connected and enabled MCP server(s). These tools are standard Genkit `ToolAction` instances and can be provided to Genkit models.\n\nMCP prompts can be fetched using `McpHost.getPrompt(serverName, promptName)` or `mcpClient.getPrompt(promptName)`. These return an `ExecutablePrompt`.\n\nAll MCP actions (tools, prompts, resources) are namespaced.\n- For `createMcpHost`, the namespace is the key you provide for that server in the `mcpServers` configuration (e.g., `localFs/read_file`).\n- For `createMcpClient`, the namespace is the `name` you provide in its options (e.g., `myFileSystemClient/list_resources`).\n\n### Tool Responses\n\nMCP tools return a `content` array as opposed to a structured response like most Genkit tools. The Genkit MCP plugin attempts to parse and coerce returned content:\n\n1. If the content is text and valid JSON, it is parsed and returned as a JSON object.\n2. If the content is text but not valid JSON, the raw text is returned.\n3. If the content contains a single non-text part (e.g., an image), that part is returned directly.\n4. If the content contains multiple or mixed parts (e.g., text and an image), the full content response array is returned.\n\n## MCP Server\n\nYou can also expose all of the tools and prompts from a Genkit instance as an MCP server using the `createMcpServer` function.\n\n```ts\nimport { googleAI } from '@genkit-ai/google-genai';\nimport { createMcpServer } from '@genkit-ai/mcp';\nimport { StdioServerTransport } from '@modelcontextprotocol/sdk/server/stdio.js';\nimport { genkit, z } from 'genkit/beta';\n\nconst ai = genkit({\n plugins: [googleAI()],\n});\n\nai.defineTool(\n {\n name: 'add',\n description: 'add two numbers together',\n inputSchema: z.object({ a: z.number(), b: z.number() }),\n outputSchema: z.number(),\n },\n async ({ a, b }) => {\n return a + b;\n }\n);\n\nai.definePrompt(\n {\n name: 'happy',\n description: 'everybody together now',\n input: {\n schema: z.object({\n action: z.string().default('clap your hands').optional(),\n }),\n },\n },\n `If you're happy and you know it, {{action}}.`\n);\n\nai.defineResource(\n {\n name: 'my resouces',\n uri: 'my://resource',\n },\n async () => {\n return {\n content: [\n {\n text: 'my resource',\n },\n ],\n };\n }\n);\n\nai.defineResource(\n {\n name: 'file',\n template: 'file://{path}',\n },\n async ({ uri }) => {\n return {\n content: [\n {\n text: `file contents for ${uri}`,\n },\n ],\n };\n }\n);\n\n// Use createMcpServer\nconst server = createMcpServer(ai, {\n name: 'example_server',\n version: '0.0.1',\n});\n// Setup (async) then starts with stdio transport by default\nserver.setup().then(async () => {\n await server.start();\n const transport = new StdioServerTransport();\n await server!.server?.connect(transport);\n});\n```\n\nThe `createMcpServer` function returns a `GenkitMcpServer` instance. The `start()` method on this instance will start an MCP server (using the stdio transport by default) that exposes all registered Genkit tools and prompts. To start the server with a different MCP transport, you can pass the transport instance to the `start()` method (e.g., `server.start(customMcpTransport)`).\n\n### `createMcpServer()` Options\n- **`name`**: (required, string) The name you want to give your server for MCP inspection.\n- **`version`**: (optional, string) The version your server will advertise to clients. Defaults to \"1.0.0\".\n\n### Known Limitations\n\n- MCP prompts are only able to take string parameters, so inputs to schemas must be objects with only string property values.\n- MCP prompts only support `user` and `model` messages. `system` messages are not supported.\n- MCP prompts only support a single \"type\" within a message so you can't mix media and text in the same message.\n\n### Testing your MCP server\n\nYou can test your MCP server using the official inspector. For example, if your server code compiled into `dist/index.js`, you could run:\n\n npx @modelcontextprotocol/inspector dist/index.js\n\nOnce you start the inspector, you can list prompts and actions and test them out manually.\n","url","https://github.com/firebase/genkit/tree/HEAD/js/plugins/mcp","githubRepository",{"_143":144,"_145":146,"_147":148,"_16":149,"_150":151,"_31":152,"_110":153,"_155":156,"_157":158,"_160":161},"defaultBranch","main","fullName","firebase/genkit","githubTreeId",1696504,948,"language","TypeScript","genkit",{"_114":115,"_106":154},"3azu1ntb9a","recentStargazers",0,"spdxLicense",{"_31":159},"Apache 2.0","stargazers",4763,"glamaJson","inspectable",true,"npmPackage",{"_31":167,"_168":169},"@genkit-ai/mcp","weeklyDownloads",1130,"supportedPlatforms","scores",{"_173":174,"_175":174,"_176":174},"license",100,"quality","security","simpleIcon","toolCount","m6x141u6rd","updatedAt","2025-10-07T19:05:25.509811Z","meta",[184,187,189,196,199,201,204,207,210,213,216,218,221,224,226,228,230,232,234],{"_185":186},"title","Genkit MCP | Glama",{"_188":93,"_31":29},"content",{"_190":191,"_192":193,"_194":195},"href","https://glama.ai/mcp/servers/@firebase/genkit","rel","canonical","tagName","link",{"_188":32,"_197":198},"property","og:title",{"_188":93,"_197":200},"og:description",{"_188":202,"_197":203},"https://glama.ai/logo.png","og:logo",{"_188":205,"_197":206},"https://glama.ai/mcp/servers/@firebase/genkit/og-image","og:image",{"_188":208,"_197":209},"630","og:image:height",{"_188":211,"_197":212},"1200","og:image:width",{"_188":214,"_197":215},"website","og:type",{"_188":191,"_197":217},"og:url",{"_188":219,"_197":220},"Glama – MCP Hosting Platform","og:site_name",{"_188":222,"_31":223},"summary_large_image","twitter:card",{"_188":219,"_31":225},"twitter:title",{"_188":93,"_31":227},"twitter:description",{"_188":205,"_31":229},"twitter:image",{"_188":208,"_31":231},"twitter:image:height",{"_188":211,"_31":233},"twitter:image:width",{"_235":236},"script:ld+json",{"_237":238,"_239":240},"@context","http://schema.org","@graph",[241],{"_237":242,"_243":244,"_245":246},"https://schema.org","@type","BreadcrumbList","itemListElement",[247,254,258,262],{"_243":248,"_249":250,"_31":251,"_252":253},"ListItem","item","/","Glama","position",1,{"_243":248,"_249":255,"_31":256,"_252":257},"/mcp","MCP",2,{"_243":248,"_249":259,"_31":260,"_252":261},"/mcp/servers","Servers",3,{"_243":248,"_249":263,"_31":32,"_252":264},"/mcp/servers/@firebase/genkit",4,"visitorIsMaintainer",false,"routes/_public/mcp/servers/~namespace/~slug/_pages/_explorer/blob/~commit/~/_route",{"_269":270,"_27":276,"_296":297,"_182":310,"_308":334,"_335":336},"blob",{"_188":271,"_272":273,"_274":275},"","fileName","architecture.svg","size",230231,{"_80":81,"_13":277,"_91":40,"_92":93,"_16":94,"_95":278,"_31":32,"_108":281,"_116":-5,"_117":118,"_119":284,"_171":295,"_177":-5,"_103":152,"_178":156,"_106":179,"_180":181},[83,84,85,86,87,88,89,90],[279],{"_98":280,"_29":105,"_106":107},{"_100":101,"_31":102,"_103":104},{"_110":282,"_103":115},{"_112":283},{"_114":115},{"_121":285,"_136":290,"_141":291,"_162":-5,"_163":164,"_165":294,"_170":-5},{"_123":286,"_133":289,"_128":135},{"_125":287,"_130":288},{"_29":127,"_128":129},{"_29":132,"_128":129},[125],{"_33":138,"_139":140},{"_143":144,"_145":146,"_147":148,"_16":149,"_150":151,"_31":152,"_110":292,"_155":156,"_157":293,"_160":161},{"_114":115,"_106":154},{"_31":159},{"_31":167,"_168":169},{"_173":174,"_175":174,"_176":174},"member",{"_298":299,"_302":303,"_304":305,"_147":306,"_307":266,"_308":273,"_309":164,"_274":275,"_128":269},"commit",{"_300":301},"sha","8e1c4c9253ae1ecbf36b22a6f8ed017511caad88","contentStoredObjectPointerId",11244053,"fullPath","py/samples/multi-server/architecture.svg",1696434,"indexable","path","safeToDisplayAsText",[311,313,315,317,319,320,321,323,324,325,326,327,328,329,330,331,332,333],{"_185":312},"architecture.svg | Genkit MCP | Glama",{"_188":314,"_31":29},"Enterprise-grade security, privacy, and features like agents, MCP, prompt templates, and more.",{"_190":316,"_192":193,"_194":195},"https://glama.ai/mcp/servers/@firebase/genkit/blob/8e1c4c9253ae1ecbf36b22a6f8ed017511caad88/py/samples/multi-server",{"_188":318,"_197":198},"architecture.svg | Genkit MCP",{"_188":314,"_197":200},{"_188":202,"_197":203},{"_188":322,"_197":206},"https://glama.ai/generated-images/og?title=architecture.svg+%7C+Genkit+MCP",{"_188":208,"_197":209},{"_188":211,"_197":212},{"_188":214,"_197":215},{"_188":316,"_197":217},{"_188":219,"_197":220},{"_188":222,"_31":223},{"_188":219,"_31":225},{"_188":314,"_31":227},{"_188":322,"_31":229},{"_188":208,"_31":231},{"_188":211,"_31":233},"py/samples/multi-server","treeMembers",[337,339,343,347],{"_298":338,"_304":305,"_308":273,"_128":269},{"_300":301},{"_298":340,"_304":341,"_308":342,"_128":269},{"_300":301},"py/samples/multi-server/multi_server.py","multi_server.py",{"_298":344,"_304":345,"_308":346,"_128":269},{"_300":301},"py/samples/multi-server/pyproject.toml","pyproject.toml",{"_298":348,"_304":349,"_308":350,"_128":269},{"_300":301},"py/samples/multi-server/README.md","README.md","actionData","errors"]

MCP directory API