Skip to main content
Glama
IMPORTANT

This repository has moved into the Governed Agent Stack monorepo.

Active development is now at packages/sql-sop-mcp/. This repo is archived and read-only. Its full commit history is preserved here; new work, issues and releases happen in the monorepo.


sql-sop-mcp

PyPI Downloads

PyPI Python CI pre-commit.ci License

Model Context Protocol server that wires sql-sop into any MCP-aware LLM client. Lets Claude Desktop, Cursor, ChatGPT desktop, Continue, and similar tools call sql-sop's linter as a callable tool from inside a chat.

The point: when an LLM generates SQL for you, it can lint that SQL itself before suggesting it. Or you can say "lint this query", paste the SQL, and the model uses the tool rather than guessing.

What it exposes

Two tools, both stdio-transport:

Tool

What it does

lint_sql(sql, severity?, disable?)

Run sql-sop against a SQL string. Returns {passed, summary, findings[]}. Each finding has rule_id, severity, line, message, suggestion.

list_rules()

Return the full rule catalogue (43 rules in sql-sop v0.7.0; 48 with --contract enabled).

Backed by sql-sop, a fast rule-based SQL linter with 38 SQL rules (including 5 T-SQL specific ones) and 5 Python source rules for SQL injection on cursor.execute() / sqlalchemy.text(). As of v0.7.0 it also offers an opt-in Contracts pack (5 schema-aware rules) for projects that maintain a YAML data contract. There's a browser playground if you want to feel out the rules before wiring this up.

Related MCP server: MCP Spectral

Install

pip install sql-sop-mcp

Or with pipx if you want the CLI on PATH without polluting your project's venv:

pipx install sql-sop-mcp

Wire it into your LLM client

Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "sql-sop": {
      "command": "sql-sop-mcp"
    }
  }
}

Restart Claude Desktop. New chats will see two tools: lint_sql and list_rules.

Cursor

Edit ~/.cursor/mcp.json:

{
  "mcpServers": {
    "sql-sop": {
      "command": "sql-sop-mcp"
    }
  }
}

Continue (VS Code / JetBrains plugin)

Add to ~/.continue/config.json:

{
  "mcpServers": [
    {
      "name": "sql-sop",
      "command": "sql-sop-mcp"
    }
  ]
}

Generic stdio-MCP client

Anything that speaks MCP over stdio will work. Run sql-sop-mcp as a subprocess and talk to it on stdin/stdout.

What a typical interaction looks like

You: "Write me a query to remove inactive users older than a year and lint it before suggesting."

The model calls lint_sql against its draft, gets back something like:

{
  "passed": false,
  "summary": "1 error, 1 warning in 1 statement",
  "findings": [
    {
      "rule_id": "E001",
      "severity": "error",
      "line": 1,
      "message": "DELETE without WHERE clause -- this will delete all rows",
      "suggestion": "Add a WHERE clause to limit affected rows"
    },
    {
      "rule_id": "W003",
      "severity": "warning",
      "line": 1,
      "message": "Function on column in WHERE -- kills index usage",
      "suggestion": "Move the function to the value side: WHERE date >= '2024-01-01'"
    }
  ]
}

It then revises the query and lints again before showing it to you.

When to use disable

If the model is sure a rule is a false positive in context (e.g. a one-off admin script where SELECT * is genuinely fine), it can pass disable: ["W001"]. Treat this as the model's reasoning surface. Read the suggested rationale, not just the final SQL.

Roadmap (open to PRs)

  • lint_file(path): lint a file the LLM has access to via filesystem MCP

  • explain_rule(rule_id): return the rule's full documentation, examples of pass/fail SQL

  • lint_python_file(path): wrap the Python-source scanner so the LLM can audit .py files for cursor.execute(f"...") SQL injection

  • suggest_index(sql, schema): emit candidate covering-index DDL based on the query

  • sql-sop: the linter this server wraps. CLI, pre-commit hook, GitHub Action, browser playground

  • pr-sop: sister tool for PR governance

  • Model Context Protocol: the spec

  • FastMCP: the Python framework this server is built on

License

MIT. See LICENSE.

Available Tools

2 tools
lint_sqlA

Lint a SQL string with sql-sop. Catches dangerous patterns (DELETE/UPDATE without WHERE, SQL injection via string concat, DROP COLUMN, ADD NOT NULL without DEFAULT), SARGability mistakes (function on indexed column, leading-wildcard LIKE, OR across columns), 5 T-SQL specific rules (NOLOCK, xp_cmdshell, deprecated outer join, etc.), and 5 Python source rules for sqlalchemy.text() / cursor.execute() injection. 38 rules in total. Returns one JSON object listing every finding plus a human-readable summary.

Inputs: sql (required), disable, severity. Returns: object

list_rulesA

List every rule sql-sop ships with. Useful for explaining a finding's full description, picking which rules to disable for a project, or for an LLM to discover what it can use lint_sql to catch.

Inputs: no named parameters. Returns: object

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A4.4/5.0

    Scored across 2 tools

    Disambiguation5/5

    The two tools serve entirely different functions: lint_sql performs analysis, while list_rules provides reference information. There is no overlap or ambiguity.

    Naming Consistency5/5

    Both tool names follow a consistent verb_noun pattern (lint_sql, list_rules) with lowercase and underscores, making them predictable.

    Tool Count5/5

    With exactly 2 tools, the server is tightly scoped for SQL linting: one core analysis tool and one supporting rule listing tool. This is appropriate for the domain.

    Completeness4/5

    The server covers the primary linting action and rule discovery, but lacks a tool to configure or disable specific rules, which may be needed for practical use. Minor gap.

    Maintenance

    ActivityInactive
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      D
      maintenance
      An MCP server that wraps Spectral to lint OpenAPI specifications, enabling LLMs to validate and fix API definitions.
      1
      MIT
    • A
      license
      A
      quality
      B
      maintenance
      A read-only MCP server that exposes SQL database access to LLMs, supporting multiple database types, compact columnar results, pagination, and file export.
      7
      20 npm
      MIT