Spark SQL MCP Server
An MCP server that enables AI assistants to query Spark SQL clusters via the Thrift/HiveServer2 protocol.
Works with any HiveServer2-compatible system: Apache Spark, AWS EMR, Hive, Impala, Presto.
接入你的智能体
發布者自己的指令。不是 PlugMCP 端點。
用我們的金鑰與日誌託管一份副本。
uvx spark-sql-mcp-server==0.1.2Claude Code 沒有本機指令的安裝連結。複製下方指令並在終端機中執行。
SPARK_HOST=<value> SPARK_PORT=<value> SPARK_DATABASE=<value> SPARK_AUTH=<value> SPARK_USERNAME=<value> SPARK_PASSWORD=<secret> SPARK_KERBEROS_SERVICE_NAME=<value> claude mcp add io.github.aidancorrell-spark-sql-mcp-server -- uvx spark-sql-mcp-server==0.1.2在終端機中執行。
{ "mcpServers": { "io.github.aidancorrell-spark-sql-mcp-server": { "command": "uvx", "args": [ "spark-sql-mcp-server==0.1.2" ], "env": { "SPARK_HOST": "<value>", "SPARK_PORT": "<value>", "SPARK_DATABASE": "<value>", "SPARK_AUTH": "<value>", "SPARK_USERNAME": "<value>", "SPARK_PASSWORD": "<secret>", "SPARK_KERBEROS_SERVICE_NAME": "<value>" } } }}claude_desktop_config.json
遠端 URL 還是本機指令? 兩種安裝方式的差異
來自發布者
An MCP server that enables AI assistants to query Spark SQL clusters via the Thrift/HiveServer2 protocol.
Works with any HiveServer2-compatible system: Apache Spark, AWS EMR, Hive, Impala, Presto.
Query Spark SQL clusters via Thrift/HiveServer2. Works with Spark, EMR, Hive, Impala.
pip install spark-sql-mcp-server
Or run directly with uvx:
uvx spark-sql-mcp-server
export SPARK_HOST="your-emr-master-node.amazonaws.com"
export SPARK_PORT="10000" # default
export SPARK_DATABASE="default" # default
export SPARK_AUTH="NONE" # NONE | LDAP | KERBEROS | CUSTOM | NOSASL
Global (all projects) — add to ~/.claude.json under your project's mcpServers:
{
"mcpServers": {
"spark-sql": {
"command": "uvx",
"args": ["spark-sql-mcp-server"],
"env": {
"SPARK_HOST": "your-emr-master-node.amazonaws.com",
"SPARK_PORT": "10000",
"SPARK_AUTH": "NONE"
}
}
}
}
Project-level — add to .claude/mcp.json in your repo:
{
"mcpServers": {
"spark-sql": {
"command": "uvx",
"args": ["spark-sql-mcp-server"],
"env": {
"SPARK_HOST": "your-emr-master-node.amazonaws.com",
"SPARK_PORT": "10000",
"SPARK_AUTH": "NONE"
}
}
}
}
Add to your claude_desktop_config.json:
{
"mcpServers": {
"spark-sql": {
"command": "uvx",
"args": ["spark-sql-mcp-server"],
"env": {
"SPARK_HOST": "your-emr-master-node.amazonaws.com",
"SPARK_PORT": "10000"
}
}
}
}
Ask Claude things like:
sales.transactions table"export SPARK_AUTH="NONE"
export SPARK_AUTH="LDAP"
export SPARK_USERNAME="your-username"
export SPARK_PASSWORD="your-password"
export SPARK_AUTH="KERBEROS"
export SPARK_KERBEROS_SERVICE_NAME="hive" # default
# Ensure you have a valid Kerberos ticket (kinit)
ssh -i your-key.pem -L 10000:localhost:10000 hadoop@your-emr-master
SPARK_HOST=localhostgit clone https://github.com/aidancorrell/spark-sql-mcp-server.git
cd spark-sql-mcp-server
pip install -e ".[dev]"
pytest
ruff check .
A Docker Compose setup provides a local Spark Thrift Server with sample data for integration testing.
# Start the Spark Thrift Server
cd docker && docker compose up -d
# Wait for it to be ready (takes ~30s on first start)
docker logs -f spark-thrift-server # look for "Sample data loaded."
# Run integration tests
pytest -m integration -v
# Tear down
cd docker && docker compose down -v
The local server comes with sample tables: default.employees, default.orders, and test_db.metrics.
Unit tests run by default with pytest (integration tests are skipped unless -m integration is specified).
With the Docker Spark server running, add it to your MCP config to test the server interactively.
Global — add to ~/.claude.json under your project's mcpServers:
{
"spark-sql": {
"command": "uvx",
"args": ["spark-sql-mcp-server"],
"env": {
"SPARK_HOST": "localhost",
"SPARK_PORT": "10000",
"SPARK_AUTH": "NONE"
}
}
}
Project-level — add to .claude/mcp.json:
{
"mcpServers": {
"spark-sql": {
"command": "uvx",
"args": ["spark-sql-mcp-server"],
"env": {
"SPARK_HOST": "localhost",
"SPARK_PORT": "10000",
"SPARK_AUTH": "NONE"
}
}
}
}
Then start a new Claude Code session and ask it to query the sample data.
The execute_query tool only allows read-only SQL statements. Queries must start with one of: SELECT, SHOW, DESCRIBE, DESC, EXPLAIN, or WITH. All other statement types (DROP, INSERT, DELETE, CREATE, ALTER, SET, ADD JAR, etc.) are rejected before reaching the Spark cluster.
Database errors are sanitized before being returned to the MCP client. Internal details such as server hostnames, file paths, and stack traces are not exposed. Connection failures report only the target host/port and error type.
SparkConfig object masks passwords in its string representationSPARK_PASSWORD is marked as a secret in the MCP registry schemaSPARK_AUTH to LDAP or KERBEROS for authenticated environments.MIT
Query Spark SQL clusters via Thrift/HiveServer2. Works with Spark, EMR, Hive, Impala.
{ "mcpServers": { "io.github.aidancorrell-spark-sql-mcp-server": { "command": "uvx", "args": [ "spark-sql-mcp-server==0.1.2" ], "env": { "SPARK_HOST": "<value>", "SPARK_PORT": "<value>", "SPARK_DATABASE": "<value>", "SPARK_AUTH": "<value>", "SPARK_USERNAME": "<value>", "SPARK_PASSWORD": "<secret>", "SPARK_KERBEROS_SERVICE_NAME": "<value>" } } }}工具
尚無工具清單。遠端伺服器會在首次健康檢查後取得工具清單。
常見問題
在本頁複製適用於 Cursor、Claude Desktop、Claude Code 或 VS Code 的 stdio 設定。它使用發布者的安裝方式,而不是 PlugMCP 代理伺服器。
發布者列出了以下環境變數:SPARK_HOST, SPARK_PORT, SPARK_DATABASE, SPARK_AUTH, SPARK_USERNAME, SPARK_PASSWORD, SPARK_KERBEROS_SERVICE_NAME。
Spark SQL 透過 stdio 在本機執行。
上次檢查時間:2026-11-09。 Query Spark SQL clusters via Thrift/HiveServer2. Works with Spark, EMR, Hive, Impala.