bedrock-guardrails-mcp-server: Add an AWS Bedrock Guardrails MCP server

[danilop]

Describe the feature

Introduce tools to call the Bedrock ApplyGuardrail API in an agentic workflow.

Two tools should be provided:

• input-only: when the content to evaluated is from a user, typically the LLM prompt.

• input-and-output: when the model output guardrails should be enforced, typically an LLM response.

Use Case

Allow an AI agent to decide when to check input or output content with configured guardrails.

Proposed Solution

Add an AWS Bedrock Guardrails MCP server (bedrock-guardrails-mcp-server).

Other Information

No response

Acknowledgements

• I may be able to implement this feature request
• This feature might incur a breaking change

[theagenticguy]

Hi @danilop just for my own understanding, could you provide an example of how this mcp server would be used in a mcp client+host such as Cline or Q Developer CLI? In other words, a conversation which uses the tools would be interrupted? Or something else?

[danilop]

Hi @laithalsaadoon – When I opened this, I was thinking broadly to any AI agent using MCP tools. Guardrails should set before and/or after the LLM activities, to check what goes in and what comes out. However, a generic AI agent (not specifically for software development such as Cline or Q Developer CLI) would benefit at different steps during an agentic workflow to be able to check that the data in input or in output to the step passes the checks provided by Bedrock Guardrails. For example, to check if an external tool returned content that includes a possible prompt attack or harmful content. If you're ok, we can leave this issue open and ask customers if they would find such a "guardrails MCP server" useful or not.

[theagenticguy]

Hi @laithalsaadoon – When I opened this, I was thinking broadly to any AI agent using MCP tools. Guardrails should set before and/or after the LLM activities, to check what goes in and what comes out. However, a generic AI agent (not specifically for software development such as Cline or Q Developer CLI) would benefit at different steps during an agentic workflow to be able to check that the data in input or in output to the step passes the checks provided by Bedrock Guardrails. For example, to check if an external tool returned content that includes a possible prompt attack or harmful content. If you're ok, we can leave this issue open and ask customers if they would find such a "guardrails MCP server" useful or not.

Makes sense! I'm curious what others think as well. What about guardrails being applied by the client?

[kenliao94]

Hey @danilop are you implementing the MCP yourself or you can take a PR from someone else? I need this use case so I can take a look as well.

[danilop]

Thanks @kenliao94 – Yes we can take a PR, see the contributing notes for how to do it efficiently.

Can you describe your use case here? I really would like to understand how it can be used.

[github-actions]

This issue is now marked as stale because it hasn't seen activity for a while. Add a comment or it will be closed soon. If you wish to exclude this issue from being marked as stale, add the "backlog" label.

[github-actions]

Closing this issue as it hasn't seen activity for a while. Please add a comment @mentioning a maintainer to reopen. If you wish to exclude this issue from being marked as stale, add the "backlog" label.