What it does
The server transpiles SQL between PostgreSQL, Oracle, Redshift, MySQL, Snowflake and Spark SQL using SQLGlot, and generates PySpark DataFrame API source text from SQL with optimization hints. It also produces AWS Glue 5.0 job template strings, including DynamicFrame conversions, Data Catalog definitions and S3 layout advice. Batch mode walks SQL files or directories and emits converted modules, while a pattern-based review flags issues in existing PySpark code and finds duplicated snippets. It generates source only: it does not start a SparkSession, submit Glue jobs or execute SQL, and optimization mode returns suggestions rather than rewritten code.
It suits migrating SQL workloads to PySpark, scaffolding Glue jobs that an agent can write to a file, and reviewing PySpark code. Recursive CTEs, MERGE, PIVOT and CONNECT BY are not converted one to one: they receive Spark SQL output with guidance. The server does not replace live catalog or plan introspection of a running Spark session. For plain transpilation, SQLGlot can be used directly.
It runs as an MCP server over stdio and is configured in clients such as Claude Desktop or Hermes Agent. The Docker image is stdio only, with no HTTP endpoint. Three tools are primary: convert, glue_job and review. Eleven older tools remain registered but are deprecated. Development requires Python 3.11 or later.
Limitations and requirements
The server can read local SQL, TXT and PDF files, and with the optional aws extra it can list and read S3 using the host's default AWS credentials. File tools are limited to the working directory or an explicit base path, which is not a sandbox. The publisher advises running it under a restricted OS account, not pointing it at secrets directories, and avoiding AWS credentials with write access unless S3 reads are intended. The Docker image supports stdio only. Development requires Python 3.11+. Eleven tools are deprecated.
Safety
Our team has not reviewed this resource. Only automated rules checked it; resources with high-risk signals are not published. These are signals, not a guarantee that installing it is safe. Try it in a sandbox project first.
- Runs code or changes an assistant's behaviour
Sources
- Official MCP Registry first seen 13 Nov 2026 · last seen 13 Nov 2026 · its page on the source
PySpark MCP
SQL migration, AWS Glue job generation, and PySpark optimization as an MCP server.
v0.1.0As shown on registry.modelcontextprotocol.io when we last read it