AI-native SQL client built on Model Context Protocol - Query databases using natural language through any LLM
sql-lens-mcp has 10 tools with consistent naming (verb_noun pattern: connect, disconnect, list_connections, execute_query, etc.). All tools have descriptions (avg ~60 chars, baseline 194). Input schemas are present with typed parameters and descriptions. However, output schemas are not documented in the source code provided, and error handling lacks recovery guidance. Parameter descriptions are minimal (10-30 chars typically), below the 72-char baseline. No tool annotations (readOnlyHint/destructiveHint) despite clear risk classifications. Missing pagination guidance for tools returning lists (describe_schema, describe_relationships). Natural language query tool (nl_query) lacks detail on LLM integration and failure modes.
Establish a connection to a database
Get foreign key relationships for a database
Get the schema information for a database
Close a database connection
Execute a SQL query against a connected database
Get the execution plan for a SQL query
List all active database connections
Execute a natural language query against a database
Get sample rows from a table
Output schemas not documented. LLMs cannot plan downstream tool calls or extract required fields (e.g., what does describe_schema return? What fields are in each table object?). Baseline: 100% of A+ tools document return types.
No tool annotations (readOnlyHint/destructiveHint/idempotentHint) despite clear risk classifications in metadata. Tools marked WRITE (connect, disconnect, execute_query) should declare destructiveHint=true to guide agent retry logic.
Parameter descriptions are minimal (10-30 chars vs 72-char baseline). E.g., 'connectionId' described as 'ID of the database connection' lacks context on format, scope, or how to obtain it. Descriptions should state WHAT, WHEN, and prerequisites.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 51 | <=2025-11-25 | v2 |
Open an interactive dashboard to view and manage all database connections
nl_query tool lacks documentation on LLM integration, model selection, and failure modes. Does it use Claude, GPT, or a local model? What happens if the LLM cannot translate the query? How are ambiguities resolved?
No pagination or result-limiting guidance for discovery tools (describe_schema, describe_relationships). If a schema has 1000 tables, does the tool return all? Baseline: tools returning lists should accept limit/offset and document max results.