Skip to content

StreamableClient tries to JSON-decode ping message #636

@ccreutzi

Description

@ccreutzi

Describe the bug
A clear and concise description of what the bug is.

If an SSE server response does not start with a message event, but a ping event, the streamable client will try to decode the ping data as JSON and fail.

I have observed this on some calls to the DeepWiki MCP server.

To Reproduce
Steps to reproduce the behaviour:

func TestStreamableClientWithDelayedResponse(t *testing.T) {
	// This test simulates a server that takes a while to respond and sends a ping before its message
	ctx := context.Background()
	fake := &fakeStreamableServer{
	t: t,
	responses: fakeResponses{
		{"POST", "", methodInitialize}: {
			header: header{
				"Content-Type":  "application/json",
				sessionIDHeader: "123",
			},
			body: jsonBody(t, initResp),
		},
		{"POST", "123", notificationInitialized}: {
			status:              http.StatusAccepted,
			wantProtocolVersion: latestProtocolVersion,
		},
		{"GET", "123", ""}: {
			header: header{
				"Content-Type": "text/event-stream",
			},
			wantProtocolVersion: latestProtocolVersion,
		},
		{"POST", "123", methodCallTool}: {
			header: header{
				"Content-Type": "text/event-stream",
			},
			wantProtocolVersion: latestProtocolVersion,
			body: `event: ping
data: ping

event: message
data: {"jsonrpc":"2.0","id":2,"result":{"content":[{"type":"text","text":"tool result"}]}}

event: ping
data: ping
`,
			},
			{"DELETE", "123", ""}: {},
		},
	}

	httpServer := httptest.NewServer(fake)
	defer httpServer.Close()

	transport := &StreamableClientTransport{Endpoint: httpServer.URL}
	client := NewClient(testImpl, nil)
	session, err := client.Connect(ctx, transport, nil)
	if err != nil {
		t.Fatalf("client.Connect() failed: %v", err)
	}

	toolAndArgs := &CallToolParams{Name: "tool", Arguments: map[string]interface{}{}}
	result, err := session.CallTool(ctx, toolAndArgs)
	if err != nil {
		t.Fatalf("client.CallTool() failed: %v", err)
	}
	expected := "tool result"
	if result == nil || len(result.Content) == 0 {
		t.Errorf("CallTool result is nil or empty")
	}
	if result.Content[0].(*TextContent).Text != expected {
		t.Errorf("CallTool result: got %q, want %q", result.Content[0].(*TextContent).Text, expected)
	}

	if err := session.Close(); err != nil {
		t.Errorf("closing session: %v", err)
	}

}

Expected behavior
A clear and concise description of what you expected to happen.

The logic taking apart an event stream should skip ping events and only tro to decode message events.

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't working

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions