serious bug: user A's notification was sent to user B

[liheng314]

Describe the bug
User A's notification was sent to user B.
I think the he problem is in method connect of protocol.ts. The local variable _transport is replaced each time connect method called.
async connect(transport: Transport): Promise<void> { this._transport = transport;

To Reproduce
Steps to reproduce the behavior:

1. create a server with tool like code below
   `
   server.tool(
   'multi-greet',
   'A tool that sends different greetings with delays between them',
   {
   name: z.string().describe('Name to greet'),
   },
   async ({ name }, { sendNotification }): Promise => {
   const sleep = (ms: number) => new Promise(resolve => setTimeout(resolve, ms));

   await sendNotification({
   method: "notifications/message",
   params: { level: "debug", data: Starting multi-greet for ${name} }
   });

   await sleep(1000); // Wait 1 second before first greeting

   await sendNotification({
   method: "notifications/message",
   params: { level: "info", data: Sending first greeting to ${name} }
   });

   await sleep(1000); // Wait another second before second greeting
   let random = Math.random()
   setInterval( async () => {
   await sendNotification({
   method: "notifications/message",
   params: { level: "info", data: Sending second greeting to ${name} ${random} }
   });
   }, 1000)

   await sleep(10000000);
   return {
   content: [
   {
   type: 'text',
   text: Good morning, ${name}!,
   }
   ],
   };
   }
   );
   `

2. run the server.

3. run a client and call the tool multi-greet

4. run another client and call the tool multi-greet

Logs
Notification received: info - Sending second greeting to MCP User 0.4445638032569461
Notification received: info - Sending second greeting to MCP User 0.2591336279269787
Notification received: info - Sending second greeting to MCP User 0.4445638032569461
Notification received: info - Sending second greeting to MCP User 0.2591336279269787
Notification received: info - Sending second greeting to MCP User 0.4445638032569461
Notification received: info - Sending second greeting to MCP User 0.2591336279269787
Notification received: info - Sending second greeting to MCP User 0.4445638032569461
Notification received: info - Sending second greeting to MCP User 0.2591336279269787

[debashish50]

You need per-client transport isolation, meaning: each connected client (tool call) should manage its own sendNotification function with its own transport.

this._transport = transport;

this._transports.set(clientId, transport);
can we do this

[liheng314]

You need per-client transport isolation, meaning: each connected client (tool call) should manage its own sendNotification function with its own transport.

this._transport = transport;

this._transports.set(clientId, transport); can we do this

yes，but i think it is the sdk itself to solve the problem not developers who use the sdk.

[ihrpr]

@liheng314 please can you share code how are you setting up clients?

[liheng314]

@liheng314 please can you share code how are you setting up clients?

@ihrpr
the latest code reports errors, but i think think the root cause is the same: multiple clients making concurrent requests will lead to some mistakes.

to reproduce:
I rewrite simpleStreamableHttp.ts code, and codes are :

import { randomUUID } from 'node:crypto';
import { McpServer } from '../../server/mcp.js';
import { StreamableHTTPServerTransport } from '../../server/streamableHttp.js';
import { z } from 'zod';
import { CallToolResult, GetPromptResult, isInitializeRequest, ReadResourceResult } from '../../types.js';
import { InMemoryEventStore } from '../shared/inMemoryEventStore.js';

// Create an MCP server with implementation details
const server = new McpServer({
  name: 'simple-streamable-http-server',
  version: '1.0.0',
}, { capabilities: { logging: {} } });

// Log the capability invocation details
server.onCapabilityChange((event) => {
  switch (event.action) {
    case 'invoked':
      console.log(`${event.capabilityType} invocation ${event.invocationIndex}: '${event.capabilityName}' started`);
      break;
    case 'completed':
      console.log(`${event.capabilityType} invocation ${event.invocationIndex}: '${event.capabilityName}' completed in ${event.durationMs}ms`);
      break;
    case 'error':
      console.log(`${event.capabilityType} invocation ${event.invocationIndex}: '${event.capabilityName}' failed in ${event.durationMs}ms: ${event.error}`);
      break;
  }
});

// Register a simple tool that returns a greeting
server.tool(
  'greet',
  'A simple greeting tool',
  {
    name: z.string().describe('Name to greet'),
  },
  async ({ name }): Promise<CallToolResult> => {
    return {
      content: [
        {
          type: 'text',
          text: `Hello, ${name}!`,
        },
      ],
    };
  }
);

// Register a tool that sends multiple greetings with notifications
server.tool(
  'multi-greet',
  'A tool that sends different greetings with delays between them',
  {
    name: z.string().describe('Name to greet'),
  },
  async ({ name }, { sendNotification }): Promise<CallToolResult> => {
    const sleep = (ms: number) => new Promise(resolve => setTimeout(resolve, ms));

    await sendNotification({
      method: "notifications/message",
      params: { level: "debug", data: `Starting multi-greet for ${name}` }
    });

    await sleep(1000); // Wait 1 second before first greeting
    let random = Math.random()
    setInterval( async () => {
      await sendNotification({
        method: "notifications/message",
        params: { level: "info", data: `Sending first greeting to ${name} ${random}` }
      });
    }, 1000)

    await sleep(1000000); // Wait another second before second greeting

    await sendNotification({
      method: "notifications/message",
      params: { level: "info", data: `Sending second greeting to ${name}` }
    });

    return {
      content: [
        {
          type: 'text',
          text: `Good morning, ${name}!`,
        }
      ],
    };
  }
);

// Register a simple prompt
server.prompt(
  'greeting-template',
  'A simple greeting prompt template',
  {
    name: z.string().describe('Name to include in greeting'),
  },
  async ({ name }): Promise<GetPromptResult> => {
    return {
      messages: [
        {
          role: 'user',
          content: {
            type: 'text',
            text: `Please greet ${name} in a friendly manner.`,
          },
        },
      ],
    };
  }
);

// Register a tool specifically for testing resumability
server.tool(
  'start-notification-stream',
  'Starts sending periodic notifications for testing resumability',
  {
    interval: z.number().describe('Interval in milliseconds between notifications').default(100),
    count: z.number().describe('Number of notifications to send (0 for 100)').default(50),
  },
  async ({ interval, count }, { sendNotification }): Promise<CallToolResult> => {
    const sleep = (ms: number) => new Promise(resolve => setTimeout(resolve, ms));
    let counter = 0;

    while (count === 0 || counter < count) {
      counter++;
      try {
        await sendNotification({
          method: "notifications/message",
          params: {
            level: "info",
            data: `Periodic notification #${counter} at ${new Date().toISOString()}`
          }
        });
      }
      catch (error) {
        console.error("Error sending notification:", error);
      }
      // Wait for the specified interval
      await sleep(interval);
    }

    return {
      content: [
        {
          type: 'text',
          text: `Started sending periodic notifications every ${interval}ms`,
        }
      ],
    };
  }
);

// Create a simple resource at a fixed URI
server.resource(
  'greeting-resource',
  'https://example.com/greetings/default',
  { mimeType: 'text/plain' },
  async (): Promise<ReadResourceResult> => {
    return {
      contents: [
        {
          uri: 'https://example.com/greetings/default',
          text: 'Hello, world!',
        },
      ],
    };
  }
);

const app = express();
app.use(express.json());

// Map to store transports by session ID
const transports: { [sessionId: string]: StreamableHTTPServerTransport } = {};

app.post('/mcp', async (req: Request, res: Response) => {
  console.log('Received MCP request:', req.body);
  try {
    // Check for existing session ID
    const sessionId = req.headers['mcp-session-id'] as string | undefined;
    let transport: StreamableHTTPServerTransport;

    if (sessionId && transports[sessionId]) {
      // Reuse existing transport
      transport = transports[sessionId];
    } else if (!sessionId && isInitializeRequest(req.body)) {
      // New initialization request
      const eventStore = new InMemoryEventStore();
      transport = new StreamableHTTPServerTransport({
        sessionIdGenerator: () => randomUUID(),
        eventStore, // Enable resumability
        onsessioninitialized: (sessionId) => {
          // Store the transport by session ID when session is initialized
          // This avoids race conditions where requests might come in before the session is stored
          console.log(`Session initialized with ID: ${sessionId}`);
          transports[sessionId] = transport;
        }
      });

      // Set up onclose handler to clean up transport when closed
      transport.onclose = () => {
        const sid = transport.sessionId;
        if (sid && transports[sid]) {
          console.log(`Transport closed for session ${sid}, removing from transports map`);
          delete transports[sid];
        }
      };

      // Connect the transport to the MCP server BEFORE handling the request
      // so responses can flow back through the same transport
      await server.connect(transport);

      await transport.handleRequest(req, res, req.body);
      return; // Already handled
    } else {
      // Invalid request - no session ID or not initialization request
      res.status(400).json({
        jsonrpc: '2.0',
        error: {
          code: -32000,
          message: 'Bad Request: No valid session ID provided',
        },
        id: null,
      });
      return;
    }

    // Handle the request with existing transport - no need to reconnect
    // The existing transport is already connected to the server
    await transport.handleRequest(req, res, req.body);
  } catch (error) {
    console.error('Error handling MCP request:', error);
    if (!res.headersSent) {
      res.status(500).json({
        jsonrpc: '2.0',
        error: {
          code: -32603,
          message: 'Internal server error',
        },
        id: null,
      });
    }
  }
});

// Handle GET requests for SSE streams (using built-in support from StreamableHTTP)
app.get('/mcp', async (req: Request, res: Response) => {
  const sessionId = req.headers['mcp-session-id'] as string | undefined;
  if (!sessionId || !transports[sessionId]) {
    res.status(400).send('Invalid or missing session ID');
    return;
  }

  // Check for Last-Event-ID header for resumability
  const lastEventId = req.headers['last-event-id'] as string | undefined;
  if (lastEventId) {
    console.log(`Client reconnecting with Last-Event-ID: ${lastEventId}`);
  } else {
    console.log(`Establishing new SSE stream for session ${sessionId}`);
  }

  const transport = transports[sessionId];
  await transport.handleRequest(req, res);
});

// Handle DELETE requests for session termination (according to MCP spec)
app.delete('/mcp', async (req: Request, res: Response) => {
  const sessionId = req.headers['mcp-session-id'] as string | undefined;
  if (!sessionId || !transports[sessionId]) {
    res.status(400).send('Invalid or missing session ID');
    return;
  }

  console.log(`Received session termination request for session ${sessionId}`);

  try {
    const transport = transports[sessionId];
    await transport.handleRequest(req, res);
  } catch (error) {
    console.error('Error handling session termination:', error);
    if (!res.headersSent) {
      res.status(500).send('Error processing session termination');
    }
  }
});

// Start the server
const PORT = 3000;
app.listen(PORT, () => {
  console.log(`MCP Streamable HTTP Server listening on port ${PORT}`);
});

// Handle server shutdown
process.on('SIGINT', async () => {
  console.log('Shutting down server...');

  // Close all active transports to properly clean up resources
  for (const sessionId in transports) {
    try {
      console.log(`Closing transport for session ${sessionId}`);
      await transports[sessionId].close();
      delete transports[sessionId];
    } catch (error) {
      console.error(`Error closing transport for session ${sessionId}:`, error);
    }
  }
  await server.close();
  console.log('Server shutdown complete');
  process.exit(0);
});

what I rewrite is code in method tool multi-greet to simulate a slow response:

    let random = Math.random()
    setInterval( async () => {
      await sendNotification({
        method: "notifications/message",
        params: { level: "info", data: `Sending first greeting to ${name} ${random}` }
      });
    }, 1000)

    await sleep(1000000);

RUN the server : npx tsx src/examples/server/simpleStreamableHttp.ts;
RUN client1: npx tsx src/examples/client/simpleStreamableHttp.ts
RUN another client2: npx tsx src/examples/client/simpleStreamableHttp.ts

then i get errors in server and client.
server error:

Error: No connection established for request ID: 1
    at StreamableHTTPServerTransport.send (/src/server/streamableHttp.ts:573:13)
    at Server.notification (src/shared/protocol.ts:607:27)
    at sendNotification (/src/shared/protocol.ts:360:16)
    at Timeout._onTimeout (/src/examples/server/simpleStreamableHttp.ts:67:13)
    at listOnTimeout (node:internal/timers:614:17)
    at process.processTimers (node:internal/timers:549:7)

client error:

Client error: Error: Failed to reconnect SSE stream: fetch failed
    at <anonymous> (/src/client/streamableHttp.ts:266:24)
    at process.processTicksAndRejections (node:internal/process/task_queues:105:5)

[Meandmybadself]

Seeing similar behavior on my end as well. Taking a deeper look into the root cause.

Understanding: Transports' session ids should set the boundaries for who a message is sent to.

[Meandmybadself]

At present, I think we need to create an new instance of our McpServer for every client connection.

If you take a peek at the Protocol.connect (which is called by McpServer.connect), you'll see there it's a singular Transport instance per Server being represented by whatever connected last.

[ihrpr]

@Meandmybadself that's right, each transport should have an instance of MCPServer, in StreambleHttpExample we use use instance of a server for a transport.

[Meandmybadself]

Thanks for the clarification, @ihrpr .

The earlier SSE example suggested we could store transports in a Map by session id, which would no longer work, correct?

[ihrpr]

@Meandmybadself it will still work, sorry, linked stateless example, stateful is here

[bhosmer-ant]

@Meandmybadself just to confirm per @ihrpr's guidance, does the intended usage pattern (one server instance per transport instance) fix this problem?

(Note: #820 should ensure that a Protocol instance won't leak responses when connect is called with client requests in flight, but this should be considered a safety measure rather than intended use.)