About SRGArticlesRecruitment Information
About SRGArticlesRecruitment Information
HOME/Articles/AWS officially releases Aurora MySQL MCP Server! Query execution in natural language

AWS officially releases Aurora MySQL MCP Server! Query execution in natural language

This is Onkai Yuta (@fat47) from the Service Reliability Group (SRG) of the Media Headquarters.
#SRG(Service Reliability Group) mainly provides cross-sectional support for the infrastructure of our media services, improving existing services, launching new ones, contributing to OSS, etc.
This article is about verifying the Aurora MySQL MCP Server, which was officially released by AWS.
I hope this helps in some way.
 

awslabs AWS MCP Servers

The official AWS MCP server group can be found here.
When it was first released there were only a few varieties, but when I looked recently there were a lot more varieties available.
 
Here are just a few recent releases to pick out some that caught my eye.

Amazon Aurora MySQL MCP Server

I'm particularly concerned aboutAmazon Aurora MySQL MCP ServerWe tried to verify the following.
The documentation for the information you need to get started can be found here.
 
Here are some features from the above document:
Features Natural language to MySQL SQL query Converting human-readable questions and commands into structured MySQL-compatible SQL queries and executing them against the configured Aurora MySQL database.
📖
function
Natural Language to MySQL SQL Query Translation - Translates human-readable questions and commands into structured, MySQL-compatible SQL queries and executes them against your configured Aurora MySQL database.
 
RDS Data API
Restrictions on using the RDS Data API
  • Not available for burstable instance classes (t3, t4)
  • Aurora MySQL version 3.07 or later is required
It is important to note that there are
 
To enable the RDS Data API on an existing cluster that meets the above requirements,
Scroll to the bottom of the Existing Cluster Connection and Security tab and change the settings that appear.
Click Enable RDS Data API.
 

Creating a verification table

Connect to Aurora MySQL and create two tables for verification.
Insert several dozen pieces of appropriate data into each table.
Sample INSERT Statement

Set up MCP with Cursor and execute queries in natural language

Here is an example using Cursor as an MCP setting.
Create .cursor/mcp.json in the directory you are working in.
 
Give instructions in the cursor chat box.
 
You will then be asked for permission to start the docker defined in mcp.json, so click “Run”.
 
テーブル一覧を取得してください
 
SHOW TABLE;
 
💡
Currently, if the query suggested above has a ; at the end, an error will occur.
This seems to be an error in the implementation of mysql-mcp-server.
https://github.com/awslabs/mcp/blob/b5b1d045a02933cf3a08e69d07d88d3722478cba/src/mysql-mcp-server/awslabs/mysql_mcp_server/server.py#L32
 
;
https://github.com/awslabs/mcp/blob/b5b1d045a02933cf3a08e69d07d88d3722478cba/src/mysql-mcp-server/awslabs/mysql_mcp_server/mutable_sql_detector.py#L101
 
;
 
You can also check the structure of a specific table.

Automatically collect slow logs and get suggestions for improvement

The slow logs will be automatically compiled and suggestions for improvement such as adding an index will be made.
 
First, as instructed, they compiled the data using mysqldumpslow.
From here, the system will automatically check the actual table situation using the MCP.
*Although the sample query only inserted a few dozen records, millions of records were added to make it a slow query.
 
The analysis results were displayed like this.
They also provide information on recommended INDEX.
 
Finally, it also lists some points to be careful about when adding an INDEX.

Risks

Being able to execute queries in natural language is very useful. In this case, the MCP server is running in read-only mode, so only read queries are allowed, making it unlikely that any major accidents will occur.
However, since the MCP server needs to be configured for each user's environment, there may be cases where it is started with read-only disabled.
As a countermeasure, it may be necessary to limit the DB user information used on the MCP server to those with appropriate permissions.
 
Also, this time we are only using it in a verification environment, but if you are using it in a production environment, it is recommended that you appropriately limit the permissions of the IAM user used by the MCP server.

Conclusion

I found that just Amazon Aurora MySQL MCP Server seemed very useful.
This time, we manually fed the slow query log to Salesforce to analyze it, but I have dreams that if we combine it with Amazon Cloudwatch Logs MCP Server, we might be able to do everything from log acquisition to query analysis in natural language!
Please try out the various official MCP servers from AWS Labs.
 
SRG is looking for people to work with us. If you are interested, please contact us here.