langchain-ai/deepagents · error · RuntimeError

Cannot reconfigure MCP session manager after sessions are ac

Error message

Cannot reconfigure MCP session manager after sessions are active

What it means

Once sessions exist, configure() only accepts connections whose signature (_connections_signature) matches the existing ones. A different signature would strand live sessions, so it raises RuntimeError.

Source

Thrown at libs/code/deepagents_code/mcp_tools.py:680

            connections: Connection configs keyed by server name.

        Raises:
            RuntimeError: If the manager is closed or reconfigured
                incompatibly after sessions already exist.
        """
        if self._closed:
            msg = "Cannot configure a closed MCP session manager"
            raise RuntimeError(msg)

        if not self._entries:
            self._connections = dict(connections)
            return

        if _connections_signature(self._connections) != _connections_signature(
            connections,
        ):
            msg = "Cannot reconfigure MCP session manager after sessions are active"
            raise RuntimeError(msg)
        self._connections = dict(connections)

    async def get_session(self, server_name: str) -> ClientSession:
        """Return a cached session for `server_name`, creating it lazily."""
        entry = self._entries.get(server_name)
        if entry is not None:
            return entry.session

        lock = self._get_lock(server_name)
        async with lock:
            entry = self._entries.get(server_name)
            if entry is not None:
                return entry.session

            entry = await self._create_entry(server_name)
            self._entries[server_name] = entry
            return entry.session

View on GitHub (pinned to a1af029e6e)

Solutions

1. Close the manager and create a new one with the new connections before opening sessions
2. Apply all config changes before the first get_session() call
3. Ensure the reconfiguration is signature-equivalent (same connections, different object identity) if that is the intent

Example fix

// before
await manager.configure(old_conns)
await manager.get_session('fs')
await manager.configure(new_conns)  # raises
// after
await manager.configure(new_conns)  # configure first
await manager.get_session('fs')

from mcp_tools import _connections_signature  # or compare your own canonical form
unchanged = _connections_signature(current) == _connections_signature(proposed)
if not unchanged:
    await manager.close(); manager = MCPSessionManager()

try:
    await manager.configure(new_connections)
except RuntimeError as e:
    if 'reconfigure' in str(e):
        await manager.close()
        manager = MCPSessionManager()
        await manager.configure(new_connections)
    else:
        raise

When it happens

Trigger: Calling configure() with a changed set of server connections while active sessions exist; passing semantically equivalent configs is allowed, differing ones (added/removed/changed servers) are not.

Common situations: Hot-reloading an MCP config file after sessions opened; adding a new MCP server at runtime without recreating the manager.

Related errors

• Invalid MCP server name {server_name!r}: token storage names
• Server '{server_name}' uses {transport!r} transport; OAuth l
• MCPServerInfo {self.name!r}: status={self.status!r} cannot c
• Cannot configure a closed MCP session manager
• Cannot create an MCP session after cleanup

AI-assisted analysis of langchain-ai/deepagents@a1af029e6e (2026-08-29). Data as JSON: /api/errors/6fec500a35843752. Report an issue: GitHub.