Agent BUilder MCP problem

In my last output the required property for MCp has been provided…

But mcp node still says error this is remaining…

I don’t understand it… It’s frustrating..

Another problem with MCP is everything all.. Still Won’t be able to execute the node …

Request id : ID cef34f25-b2c7-4609-aee9-c9ab8ae4ff65

I’m fairly certain the MCP node just doesn’t work right now.

If you add an MPC server under My Agent in tools it will…but I’ve been fighting the node itself for hours.

The furthest I got was no errors and then when trying in the preview: Workflow failed: Reauthentication required. (code: server_error)

Having exactly the same issue. Seems like the “stand alone” MCP component is not yet baked.

Have you been able to figure out how to pass the input from the AGENT to the MCP Server? Im doing a basic add_numbers(a, b) MCP call, and I receive the same error as you. Ive tried many options in defining the output scheme from the agent without any luck:
AGENT → Json output schema with a and b → MCP SERVER add_numbers(a, b) function.

I also noticed that only 1 function can be assigned:

The MCP Server works with the Chat (Responses), so I know its an Agent Builder (ChatKit) problem only…

Hi @che.kulhan

I think I see what’s going on here.

This looks like a mismatch between how the Agent structures its output and what the MCP tool is expecting. Even if you define {a, b} in the structured output, the agent usually nests that under output_parsed, so a and b don’t end up at the top level.

That’s why the MCP step is saying those fields are missing.

One thing you can try is adding a Transform step between the Agent and MCP to remap things, like:

a = output_parsed.a  

b = output_parsed.b

That should flatten it so it lines up with the MCP input schema.

You can refer to this Doc for more information.

Curious if that fixes it on your side.

~Smith

Here's the doc link as mentioned above

Thanks for that. I was successfully able to pass the input data from the Agent to the MCP Server via a Transform object.

I noticed that only one MCP tool can be called when it is done this way using the MCP component. One cannot select multiple function calls…A little limiting, in my opinion.

In addition, another Agent must be used after the MCP Tool to display the result. Seems like a lot of components required, just to display a simple tools call.

I have built and hosted this demo MCP Sever using AWS Bedrock Agent Code tools, so good to see that these can be integrated into Agent Builder for future projects.