feat(MCP Client Tool Node): Add MCP Client Tool Node to connect to MCP servers over SSE (#14464)

Co-authored-by: कारतोफ्फेलस्क्रिप्ट™ <aditya@netroy.in>
Co-authored-by: JP van Oosten <jp@n8n.io>
This commit is contained in:
Elias Meire
2025-04-09 17:31:53 +02:00
committed by GitHub
parent b52f9f0f6c
commit 34252f53f9
24 changed files with 926 additions and 35 deletions

View File

@@ -0,0 +1,173 @@
import { WebhookAuthorizationError } from 'n8n-nodes-base/dist/nodes/Webhook/error';
import { validateWebhookAuthentication } from 'n8n-nodes-base/dist/nodes/Webhook/utils';
import type { INodeTypeDescription, IWebhookFunctions, IWebhookResponseData } from 'n8n-workflow';
import { NodeConnectionTypes, Node } from 'n8n-workflow';
import { getConnectedTools } from '@utils/helpers';
import type { CompressionResponse } from './FlushingSSEServerTransport';
import { McpServerSingleton } from './McpServer';
import type { McpServer } from './McpServer';
const MCP_SSE_SETUP_PATH = 'sse';
const MCP_SSE_MESSAGES_PATH = 'messages';
export class McpTrigger extends Node {
description: INodeTypeDescription = {
displayName: 'MCP Server Trigger',
name: 'mcpTrigger',
icon: {
light: 'file:../mcp.svg',
dark: 'file:../mcp.dark.svg',
},
group: ['trigger'],
version: 1,
description: 'Expose n8n tools as an MCP Server endpoint',
activationMessage: 'You can now connect your MCP Clients to the SSE URL.',
defaults: {
name: 'MCP Server Trigger',
},
codex: {
categories: ['AI', 'Core Nodes'],
subcategories: {
AI: ['Root Nodes', 'Model Context Protocol'],
'Core Nodes': ['Other Trigger Nodes'],
},
alias: ['Model Context Protocol', 'MCP Server'],
resources: {
primaryDocumentation: [
{
url: 'https://docs.n8n.io/integrations/builtin/core-nodes/n8n-nodes-langchain.mcptrigger/',
},
],
},
},
triggerPanel: {
header: 'Listen for MCP events',
executionsHelp: {
inactive:
"This trigger has two modes: test and production.<br /><br /><b>Use test mode while you build your workflow</b>. Click the 'test step' button, then make an MCP request to the test URL. The executions will show up in the editor.<br /><br /><b>Use production mode to run your workflow automatically</b>. <a data-key='activate'>Activate</a> the workflow, then make requests to the production URL. These executions will show up in the <a data-key='executions'>executions list</a>, but not the editor.",
active:
"This trigger has two modes: test and production.<br /><br /><b>Use test mode while you build your workflow</b>. Click the 'test step' button, then make an MCP request to the test URL. The executions will show up in the editor.<br /><br /><b>Use production mode to run your workflow automatically</b>. Since your workflow is activated, you can make requests to the production URL. These executions will show up in the <a data-key='executions'>executions list</a>, but not the editor.",
},
activationHint:
'Once youve finished building your workflow, run it without having to click this button by using the production URL.',
},
inputs: [
{
type: NodeConnectionTypes.AiTool,
displayName: 'Tools',
},
],
outputs: [],
credentials: [
{
// eslint-disable-next-line n8n-nodes-base/node-class-description-credentials-name-unsuffixed
name: 'httpBearerAuth',
required: true,
displayOptions: {
show: {
authentication: ['bearerAuth'],
},
},
},
{
name: 'httpHeaderAuth',
required: true,
displayOptions: {
show: {
authentication: ['headerAuth'],
},
},
},
],
properties: [
{
displayName: 'Authentication',
name: 'authentication',
type: 'options',
options: [
{ name: 'None', value: 'none' },
{ name: 'Bearer Auth', value: 'bearerAuth' },
{ name: 'Header Auth', value: 'headerAuth' },
],
default: 'none',
description: 'The way to authenticate',
},
{
displayName: 'Path',
name: 'path',
type: 'string',
default: '',
placeholder: 'webhook',
required: true,
description: 'The base path for this MCP server',
},
],
webhooks: [
{
name: 'setup',
httpMethod: 'GET',
responseMode: 'onReceived',
isFullPath: true,
path: `={{$parameter["path"]}}/${MCP_SSE_SETUP_PATH}`,
nodeType: 'mcp',
ndvHideMethod: true,
ndvHideUrl: false,
},
{
name: 'default',
httpMethod: 'POST',
responseMode: 'onReceived',
isFullPath: true,
path: `={{$parameter["path"]}}/${MCP_SSE_MESSAGES_PATH}`,
nodeType: 'mcp',
ndvHideMethod: true,
ndvHideUrl: true,
},
],
};
async webhook(context: IWebhookFunctions): Promise<IWebhookResponseData> {
const webhookName = context.getWebhookName();
const req = context.getRequestObject();
const resp = context.getResponseObject() as unknown as CompressionResponse;
try {
await validateWebhookAuthentication(context, 'authentication');
} catch (error) {
if (error instanceof WebhookAuthorizationError) {
resp.writeHead(error.responseCode);
resp.end(error.message);
return { noWebhookResponse: true };
}
throw error;
}
const mcpServer: McpServer = McpServerSingleton.instance(context.logger);
if (webhookName === 'setup') {
// Sets up the transport and opens the long-lived connection. This resp
// will stay streaming, and is the channel that sends the events
const postUrl = req.path.replace(
new RegExp(`/${MCP_SSE_SETUP_PATH}$`),
`/${MCP_SSE_MESSAGES_PATH}`,
);
await mcpServer.connectTransport(postUrl, resp);
return { noWebhookResponse: true };
} else if (webhookName === 'default') {
// This is the command-channel, and is actually executing the tools. This
// sends the response back through the long-lived connection setup in the
// 'setup' call
const connectedTools = await getConnectedTools(context, true);
const wasToolCall = await mcpServer.handlePostMessage(req, resp, connectedTools);
if (wasToolCall) return { noWebhookResponse: true, workflowData: [[{ json: {} }]] };
return { noWebhookResponse: true };
}
return { workflowData: [[{ json: {} }]] };
}
}