Bedrock Prompts

You run the Bedrock Prompts MCP Server to manage and invoke Bedrock managed prompts from a local or cloud environment. This server lets you browse prompts, view details, run prompts with variable substitutions, stream results, and perform batch invocations across multiple inputs, all while supporting multiple model families through the Bedrock ecosystem.

How to use

Connect with a MCP client to interact with Bedrock prompts using natural language or structured requests. You can list available prompts, fetch detailed configurations, invoke prompts with specific variables, stream real-time outputs when supported, run multiple invocations in parallel, and inspect different versions of a prompt.

How to install

Prerequisites: you need Python 3.10 or higher, AWS credentials configured, access to AWS Bedrock, and (optionally) Claude Desktop for MCP integration.

Option 1: Local Development

# Clone or download the files
cd /path/to/bedrock-prompts-mcp

# Install dependencies
pip install -r requirements.txt

# Test the server
python bedrock_prompts_mcp_server.py

Option 2: Install as Package

pip install -e .

AWS configuration and environment

Ensure your AWS credentials are configured. The server uses the default credential chain, so you can set them via the AWS CLI or environment variables.

# Via AWS CLI
aws configure

# Or set environment variables
export AWS_ACCESS_KEY_ID=your_key
export AWS_SECRET_ACCESS_KEY=your_secret
export AWS_REGION=us-east-1

Optional environment variables include AWS_REGION (default us-east-1) and BEDROCK_TENANT_ID for tenant scoping.

Claude Desktop configuration (MCP integration)

Configure Claude Desktop to load the Bedrock Prompts MCP server and provide AWS region context.

{
  "mcpServers": {
    "bedrock-prompts": {
      "command": "python",
      "args": ["/absolute/path/to/bedrock_prompts_mcp_server.py"],
      "env": {
        "AWS_REGION": "us-east-1"
      }
    }
  }
}

Using the package form with Claude Desktop

If you installed as a package, use the MCP entry as your server reference.

{
  "mcpServers": {
    "bedrock-prompts": {
      "command": "bedrock-prompts-mcp",
      "env": {
        "AWS_REGION": "us-east-1"
      }
    }
  }
}

Usage examples in Claude Desktop

Once configured, you can interact with Bedrock prompts using natural language. Examples include listing prompts, retrieving details for a specific prompt, invoking a prompt with questions, batch invocations with different inputs, and streaming responses when available.

Troubleshooting

If the server won’t start, verify your AWS credentials, ensure Python is version 3.10 or newer, and check the logs in Claude Desktop for errors.

If you see a "No credentials found" error, run aws configure or provide credentials via environment variables.

If prompt invocation fails, confirm the prompt exists, the targeted model is available in your region, and you have permission to invoke Bedrock models.

If variable substitution does not work, ensure your variable names match exactly and that the prompt expects those variables.

Security notes

Never commit AWS credentials to version control. Prefer IAM roles where possible and follow least-privilege principles for permissions. The server is designed to run locally without exposing network endpoints.

Required IAM permissions

To use Bedrock prompts via MCP, your IAM user or role should include permissions to list prompts, get prompt details, list prompt versions, and invoke models.

Available tools

list_bedrock_prompts

Lists all Bedrock managed prompts available for your account.

get_bedrock_prompt_details

Retrieves detailed information about a specific Bedrock prompt, including templates and variables.

invoke_bedrock_prompt

Invokes a Bedrock prompt with provided variables and returns the model's response.

list_bedrock_prompt_versions

Lists all versions of a specific Bedrock prompt.

invoke_bedrock_prompt_stream

Invokes a Bedrock prompt with streaming responses for real-time output.

batch_invoke_bedrock_prompt

Invokes a Bedrock prompt multiple times with different variable sets in parallel.