# csvql MCP server

> Allows AI agents to run read-only SQL queries directly on CSV and Excel files without data ingestion or network access.

- Repository: https://github.com/melihbirim/csvql
- Page: https://hostedmcp.org/servers/melihbirim/csvql
- Transport: stdio
- Stars: 31
- License: MIT
- Latest version: 3.1.0

## FAQ

### Is csvql a database?

No, it is a query engine that runs SQL on existing CSV, TSV, or Excel files without requiring data import or ingestion.

### Can the AI agent modify my files?

No, the server is read-only by construction and the parser does not support write statements like INSERT, UPDATE, or DELETE.

### How does csvql handle large files without crashing the AI context?

It uses bounded output, capping MCP results at approximately 100 rows or 12 KB per call to prevent flooding the context.

### Can I restrict access to specific sensitive columns?

Yes, you can use the --deny-columns flag to prevent the agent from reading specific columns like email or SSN.

### Does it require an internet connection to work?

No, the binary contains no network code, making it safe for private data environments.
