[BUG] MCP Client does not propagate contextvars

[somaraani]

Checks

• I have updated to the lastest minor and patch version of Strands
• I have checked the documentation and this is not expected behavior
• I have searched ./issues and there are no duplicates of my issue

Strands Version

1.21.1

Python Version

3.10

Operating System

MacOS 26.2

Installation Method

pip

Steps to Reproduce

# Factory function runs in background thread
mcp_client = MCPClient(lambda: streamablehttp_client(MCP_URL))

# Set Context Var in Current thread
context_var.set("some-value")

# The factory runs in BACKGROUND thread, so current contextvars do not propagate
# This makes then inaccessible to transports or auth logic that requires them
with mcp_client:
    # Make request requiring context var

Expected Behavior

Contextvar propagated to background thread and is accessible by transports.

Actual Behavior

New thread is created without propagating current contextvars

Additional Context

In MCPClient.start, the background thread is started with

 self._background_thread = threading.Thread(target=self._background_task, args=[], daemon=True)

This issue was fixed in #1146 for tool invocations by propagating copying contextvars in the ThreadPoolExecutor.

Possible Solution

This can be fixed by copying context vars in the start method when starting the background thread.

# Copy the current context
ctx = contextvars.copy_context()

# Create the thread with the context
self._background_thread = threading.Thread(
    target=ctx.run,
    args=(self._background_task,),
    daemon=True
)

Related Issues

#946, #946

[cagataycali]

🤖 Analysis by strands-coder autonomous agent

Root Cause Confirmed

I've analyzed this issue and confirmed the bug. As you correctly identified:

Problem: In MCPClient.start() (line ~182), the background thread is created without propagating context variables:

self._background_thread = threading.Thread(target=self._background_task, args=[], daemon=True)

Similar Fix Already Applied

This exact issue was fixed for the ThreadPoolExecutor in PR #1146 (src/strands/_async.py). The pattern used there is:

context = contextvars.copy_context()
future = executor.submit(context.run, execute)

Proposed Solution

Apply the same pattern to MCPClient.start():

# Copy the current context
ctx = contextvars.copy_context()

# Create the thread with the context
self._background_thread = threading.Thread(
    target=ctx.run,
    args=(self._background_task,),
    daemon=True
)

Changes Required

1. Add import contextvars at the top of mcp_client.py (if not already there)
2. Modify the start() method to copy and run context

I'll Create a PR

I'm going to create a PR with this fix. The change is minimal and follows the exact same pattern already established in the codebase.

---

Analysis by strands-coder autonomous agent 🦆

[cagataycali]

✅ PR Created: #1443

I've submitted a fix following your suggested solution. The change is minimal and follows the exact pattern already established in PR #1146.

Please review when you have a chance! 🦆

[cagataycali]

Hi @somaraani! 👋

Great catch on this bug! I've created a PR to fix this:

PR #1444: #1444

The fix follows the exact approach you suggested - using contextvars.copy_context() and ctx.run() when creating the background thread. This is consistent with how PR #1146 solved the same issue for tool invocations.

Changes made:

1. Added contextvars import
2. Modified start() method to copy context before creating background thread
3. Added a test to verify context propagation

All existing MCP tests (50 tests) pass, plus the new contextvar test.

Thanks for the detailed issue report and the suggested solution! 🦆

---

strands-coder autonomous agent