Bug

The longRunningOperation tool in examples/everything causes a transport error: transport closed when called over stdio, even with minimal parameters (duration: 0.1, steps: 1).

Reproduction

cd examples/everything
go build -o /tmp/everything-server .

# Call the tool via MCP stdio
echo '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"test","version":"1.0"}}}' | timeout 10 /tmp/everything-server

Or using mcp-assert:

name: longRunningOperation completes
server:
  command: /tmp/everything-server
assert:
  tool: longRunningOperation
  args:
    duration: 0.1
    steps: 1
  expect:
    not_error: true
timeout: 30s

FAIL  longRunningOperation completes                                  152ms
      tool call longRunningOperation failed: transport error: transport closed

Root cause

The tool handler at main.go:418-459 calls time.Sleep() during execution. When progressToken is nil (no token in request meta), the tool still sleeps. The stdio transport appears to close before the response is sent, causing a race between the sleep completing and the transport teardown.

The other 5 tools in the same server (echo, add, notify, getTinyImage, get_resource_link) all pass without issue.

Environment

• mcp-go: latest (cloned from HEAD)
• Go: 1.25.5
• OS: Ubuntu (GitHub Actions CI) and macOS
• Transport: stdio
• Client: mcp-assert v0.1.1

Found while building a comprehensive test suite for the mcp-go SDK example servers. All other tools in everything, typed_tools, and structured_input_and_output pass (17/18 assertions).