redshift-mcp-server: Transaction Aborted Error

[robertowtr-circle]

Describe the bug

We are encountering issues when using the Redshift MCP through Claude and Visual Studio Code. The MCP successfully connects and lists clusters and databases, but after running a query that fails (for example, due to an invalid SQL statement), subsequent queries consistently fail with the following error:
Error executing tool execute_query: Statement failed: ERROR: current transaction is aborted, commands ignored until end of transaction block
It appears that once a query execution fails, the session or transaction remains in an aborted state, and all following queries return this error until the MCP connection is reset.

Expected Behavior

After an individual query error, the MCP should handle the failure gracefully and continue to execute future valid queries without requiring a reset.

Current Behavior

When the Redshift MCP executes a query that fails (for example, due to invalid SQL syntax or referencing a non-existent table), the session appears to enter a persistent error state. After this initial failure, every subsequent query attempt returns the same error.

Reproduction Steps

1.	Connect to Redshift MCP from Claude or Visual Studio Code.
2.	Verify connectivity by running a valid query, for example: 

SELECT 1;
→ Query executes successfully.
3. Run an invalid query, for example:
SELECT * FROM non_existent_table;
→ Query fails as expected (due to invalid table reference).
4. Immediately try running another valid query, such as:
SELECT 1;
→ Observe that the MCP now returns the following error:
Statement failed: ERROR: current transaction is aborted, commands ignored until end of transaction block

Possible Solution

No response

Additional Information/Context

No response

OS

Mac

Server

other

Server Version

No response

Region experiencing the issue

us-east-1

Other information

No response

Service quota

• I have reviewed the service quotas for this construct

[maanpthk]

Yes, you're correct, I have also been facing the same issue for the last 2 weeks in my application! After finally taking a look at recent changes in the codebase I think this issue arose after a recent change where execute statement with session was implemented. The problem might be occurring because after a FAILED query, the function tries to END the transaction without executing ROLLBACK first. Since Redshift puts failed transactions into an aborted state, the END statement fails, leaving the session corrupted.
I have raised a PR(#1495) to fix this by properly executing ROLLBACK; instead of END; when a query fails within a transaction block..

Temporary workaround: Until this is fixed, you can use version 0.0.7 which doesn't have this issue:
use awslabs.core-mcp-server@0.0.7 in args. instead of awslabs.core-mcp-server@latest

[grayhemp]

Thank. you very much for the detailed report, @robertowtr-circle. Will look at it with priority.

[robertowtr-circle]

Thank you, @maanpthk — this workaround did the trick!
And thanks as well, @grayhemp . Much appreciated!

[grayhemp]

This is weird why we see the error. Because

dev=> begin;
BEGIN
dev=*> select aaa;
ERROR:  column "aaa" does not exist
dev=!> select 1;
ERROR:  current transaction is aborted, commands ignored until end of transaction block
dev=!> end;
ROLLBACK
dev=> select 1;
 ?column? 
----------
        1
(1 row)

END should work as both COMMIT is success and ROLLBACK if not.

Let me dig into that. Standby...

[grayhemp]

Oh, I think I got it. The code never gets to executing END because of the exception. Should be an easy fix. Thanks again for reporting.

[grayhemp]

@maanpthk @robertowtr-circle I created another PR which fixes the issue #1512. Please let me know if you have questions.