MCP does not work with certain valid MariaDB identifiers

[jbdocuphase]

Steps to reproduce:

1. Create a table called foo.bar
2. Ask agent to get the schema for the table foo.bar
3. MCP will error saying the table identifier is invalid.

Unfortunately, if you ask the agent to quote the identifier, the MCP call will still error.

Error stems from the fact that the isidentifier Python string method is being utilized instead of using a different method to enforce MariaDB's identifier rules.

[coolaj86]

edit: Unfortunately that patch doesn't actually fix the problem for me with my-db-name . If I add the backticks to the ENVs it fails on startup, otherwise it still fails.

How to ad-hoc add this patch:

git clone https://github.com/MariaDB/mcp ./mariadb-mcp
pushd ./mariadb-mcp/

git switch -c fix-mariadb-identifiers
git pull https://github.com/jbdocuphase/mcp main

[jbdocuphase]

edit: Unfortunately that patch doesn't actually fix the problem for me with my-db-name . If I add the backticks to the ENVs it fails on startup, otherwise it still fails.

How to ad-hoc add this patch:

git clone https://github.com/MariaDB/mcp ./mariadb-mcp
pushd ./mariadb-mcp/

git switch -c fix-mariadb-identifiers
git pull https://github.com/jbdocuphase/mcp main

Hey @coolaj86, sorry I left a comment on the PR that it needed some updates. However, I just pushed those updates up, however, for clarification you're saying the MCP server is failing to start-up if the database you're trying to connect to has special characters? My changes were restricted to the MCP endpoints themselves.

Out of curiosity, can you post the error you're getting?

[jacekfe]

I can confirm the problem.
For the databases with . (dot) in the name all operations fail.
I.e. list_tables
Error executing tool list_tables: Invalid database name provided: blog.xxx

For databases without dots in the name list_tables works fine.

[jbdocuphase]

@jacekfe thanks for confirming. Would you be able to pull the PR #23 and verify it resolves your problem?

[itissdr]

Found and fixed this problem before I saw the issue posted here. Just finished a pull request. #48

[jbdocuphase]

@itissdr I had fixed this in PR #23, I checked your PR and it's very similar to my first attempt, however, this approach doesn't fix all cases.

I ended up taking a different approach to normalize the identifiers and deal with edge cases like identifiers that are encapsulated in backticks. e.g. `foo.bar` or not foo.bar

This also means I had to update the CREATE/DROP/INSERT, etc queries to accommodate these changes.

I hope mine gets merged at some point, it's been 6 months, no feedback from maintainers and at this point probably dying of PR bit rot.

[itissdr]

@itissdr I had fixed this in PR #23, I checked your PR and it's very similar to my first attempt, however, this approach doesn't fix all cases.

I ended up taking a different approach to normalize the identifiers and deal with edge cases like identifiers that are encapsulated in backticks. e.g. `foo.bar` or not foo.bar

This also means I had to update the CREATE/DROP/INSERT, etc queries to accommodate these changes.

I hope mine gets merged at some point, it's been 6 months, no feedback from maintainers and at this point probably dying of PR bit rot.

Thanks for the feedback, my specific issue was the kebab case so I fixed my immediate problem before it even occured to me check if anyone else had seen it.