All notable changes to this project will be documented in this file.
The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.
- Add MCPClient and via to user agent (#1724)
- Throw error instead of return response in MCP tools (#1704)
- Validate origin/host headers in streamable-http mode (#1683)
- Upgrade AWS CLI to v1.42.65 (#1646)
- Change priorities of default_region parameters in call_aws_helper (#1641)
- Upgrade AWS CLI to v1.42.64 (#1637)
- Add region parameter support to call_aws_helper (#1622)
- Upgrade AWS CLI to v1.42.57 (#1573)
- Log errors thrown by the agent scripts manager (#1533)
- Converted MCP server to use FastMCP framework instead of python mcp sdk (#1513)
- Add call_aws helper function to consume credentials from other sources (#1547)
- CLI commands that don't expect any parameters (#1494)
- Support for
--endpoint-urlflag for localhost endpoints (#1452) - Change max retries to 3 when interpreting CLI command (#1485)
- Agent Script for creating Aurora DB with instances (#1401)
- AWS_API_MCP_STATELESS_HTTP configuration option (#1349)
- Replace local knowledge base with a remote endpoint for
suggest_aws_commands(#1282)
waitand other polling AWS CLI commands(#1402)
- Command output logging (#1388)
- Mark more operations as mutating (#1387)
- Mark sts:AssumeRole as mutating (#1364)
- Agent script for CloudTrail Multi-Region Setup (#1320)
- Add telemetry for AWS CLI customizations (#1335)
- Enforcement of
AUTH_TYPE=no-authfor streamable-http mode (#1345) - Agent Script for troubleshooting permissions using CloudTrail events (#1313)
- Loading of security policy from
~/.aws/aws-api-mcp/mcp-security-policy.json(#1311) - Enforcement of
READ_OPERATIONS_ONLY_MODEandREQUIRE_MUTATION_CONSENTin security policy (#1301)
- Agent Script for debugging Lambda timeouts (#1271)
- Agent Script for failure troubleshooting (#1276)
- Safe execution for AWS APIs within working directory (#1261)
- Support for custom security policy configuration via
~/.aws/aws-api-mcp/mcp-security-policy.jsonfile (#1213) - Custom deny list and elicitation required lists for AWS operations with pattern matching support (#1213)
- Documentation for security policy configuration in README (#1240)
- Support for custom agent scripts directory via
AWS_API_MCP_AGENT_SCRIPTS_DIRenvironment variable (#1227) - Scrubbing of sensitive logs (#1228)
- Telemetry for consent mechanism (#1202)
- Support for streamable HTTP transport mode via
AWS_API_MCP_TRANSPORTenvironment variable (#1192) - Configurable port for HTTP transport mode via
AWS_API_MCP_PORTenvironment variable (defaults to 8000) (#1192) - Configurable host for HTTP transport mode via
AWS_API_MCP_HOSTenvironment variable (defaults to 127.0.0.1) (#1192)
- Support commands with outfile parameter (#1154)
- Experimental support for Agent Scripts (#1149)
- Fetch embedding model from AWS instead of Hugging Face (#1127)
- Use region from profile specified in cli command (#1123)
- Validate
AWS_REGIONenvironment variable (#1030)
- Async model loading on Windows (#1035)
- Improve tool logging (#1004)
- Update README (#1020)
- Support for
--profilein boto3 operations. (#986)
- First version of the consent mechanism using elicitation. This can be enabled using
REQUIRE_MUTATION_CONSENTand will prompt for input before executing any mutating operations. (#926)
- Load the sentence transformers in the background (#844)
- Switched to CPU-only PyTorch (#856)
- Tool annotations (#915)
AWS_REGIONis no longer a mandatory environment variable. The region will be determined similar to boto3 with a default fallback tous-east-1(#952)
- Support profile for customizations (e.g.
s3 ls) (#896)
- First release of AWS API MCP Server.
call_awstool. Executes AWS CLI commands with validation and proper error handlingsuggest_aws_commandstool. Suggests AWS CLI commands based on a natural language query. This tool helps the model generate CLI commands by providing a description and the complete set of parameters for the 5 most likely CLI commands for the given query, including the most recent AWS CLI commands - some of which may be otherwise unknown to the model.