@arkty/redash-mcp

MCP Server for Redash — execute SQL queries, explore database schemas, and get results as JSON or CSV.

• Ad-hoc queries without creating objects in Redash (no trace left)
• SQL write guard — destructive operations (INSERT/UPDATE/DELETE/DROP) are blocked
• CSV export — saves results to file, returns summary to LLM
• Schema exploration — list tables, search columns, inspect structure
• Default data source — configure once, query without specifying every time

Installation

npx -y @arkty/redash-mcp

Configuration

Claude Code (.mcp.json)

{
  "mcpServers": {
    "redash": {
      "command": "npx",
      "args": ["-y", "@arkty/redash-mcp"],
      "env": {
        "REDASH_URL": "https://your-redash.example.com",
        "REDASH_API_KEY": "your-api-key",
        "REDASH_DEFAULT_DATA_SOURCE_ID": "1"
      }
    }
  }
}

Environment Variables

Variable	Required	Default	Description
REDASH_URL	Yes	—	Redash instance URL
REDASH_API_KEY	Yes	—	User API key (from Redash profile settings)
REDASH_DEFAULT_DATA_SOURCE_ID	No	—	Default data source ID
REDASH_DEFAULT_TABLE	No	—	Default table name (shown in tool description)
REDASH_TIMEOUT	No	30000	Request timeout in ms
REDASH_MAX_ROWS	No	500	Max rows in JSON response
REDASH_POLL_INTERVAL	No	1000	Job polling interval in ms
REDASH_MAX_POLL_ATTEMPTS	No	120	Max poll attempts before timeout
REDASH_OUTPUT_DIR	No	.	Directory for CSV file output
REDASH_EXTRA_HEADERS	No	—	Extra HTTP headers (JSON or key=value;key2=value2)

Tools

execute_adhoc_query

Execute SQL directly against a data source. No query object is created in Redash. Destructive SQL is blocked.

Parameters: query, data_source_id?, format? ("json" | "csv")

list_data_sources

List all available data sources with IDs, names, and types.

get_data_source_schema

Get full database schema (all tables and columns) for a data source.

Parameters: data_source_id

get_table_info

Get column names and types for a specific table.

Parameters: data_source_id, table_name

search_schema

Search tables and columns by name pattern (case-insensitive).

Parameters: data_source_id, search

Security

• Destructive SQL (INSERT, UPDATE, DELETE, DROP, ALTER, TRUNCATE, CREATE, etc.) is blocked at the MCP server level. The filter is intentionally simple — it checks for keyword presence in the query text. This means it may produce false positives (e.g., SELECT REPLACE(...) is blocked because REPLACE is a keyword). This is a safety net, not a bulletproof solution.
• Use read-only database credentials in your Redash data source configuration — this is the real protection layer
• API key is passed via environment variable, never hardcoded

License

MIT