A collection of MCP servers for text-to-SQL query generation, database execution, RAG-based document retrieval, and image text extraction using AWS Bedrock and LangChain
This server suite exhibits critical definition quality gaps across all 8 tools. Most tools lack proper parameter descriptions (only 'question' and 'sql' have minimal descriptions; 'code', 'keyword', 'image_base64/image_path/prompt', 'format' lack descriptions entirely). Output schemas are completely undocumented, no tool declares what it returns or what fields the LLM should expect. Tool naming is inconsistent: 'generate_query' appears twice (tools 1 and 3), 'execute_query' vs 'execute_athena_query' creates ambiguity. The 'execute_code' tool is dangerously under-specified with no constraints on what code can run. Error handling is absent, no guidance on recovery, retryability, or what to do on failure. Descriptions are generic and under 50 chars for most tools, violating the 10 - 1024 char baseline. No tool declares idempotency, destructiveness, or read-only status in the schema itself (only in the Risk field, which is not part of the MCP protocol).
Execute SQL against the database and return the relevant information. sql: SQL statement to execute return: Result of the SQL execution
Execute Python code and return stdout/stderr output. Use this tool to run Python code for tasks such as processing data, processing data, or performing computations. The execution environment has access to common libraries: pandas, numpy, matplotlib, seaborn, etc. json, csv, os, requests, etc.
Execute SQL against the database and return the relevant information. sql: SQL statement to execute return: Result of the SQL execution
Extract text from an image using an LLM (AWS Bedrock). Provide either image_base64 or image_path. The image is resized automatically if it exceeds size limits (5MB, ~2M pixels).
Generate an SQL statement to query the database. question: Information to retrieve from the database return: Generated SQL statement
Generate an SQL statement to query the database. question: Information to retrieve from the database return: Generated SQL statement
Duplicate tool names: 'generate_query' defined in both mcp_server_athena.py and mcp_server_text2sql.py. LLMs cannot distinguish between them; both will be registered with identical names, causing routing ambiguity.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields. E.g., does 'generate_query' return a string or an object with {sql, explanation}? Does 'execute_query' return rows, count, or raw JSON?
'execute_code' tool has no input validation, no constraints on code execution, no timeout, and no description of what libraries are available. Agents can execute arbitrary Python, including file I/O, network calls, and system commands. This is a critical security and safety gap.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 37 | <=2025-11-25 | v2 |
Returns the current date and time in Asia/Seoul, including the Korean weekday. Example: "2026-08-08 15:51:06 (토요일)"
Query the keyword using RAG based on the knowledge base. keyword: the keyword to query return: the result of query
Parameter descriptions are missing or trivial. 'question', 'sql', 'keyword', 'code', 'image_base64', 'image_path', 'prompt', 'format' lack actionable guidance. E.g., 'code' has no description at all; LLMs cannot infer what code is safe to run or what libraries are available.
No error handling guidance. Tools do not document what errors can occur, whether they are retryable, or what the LLM should do next. E.g., if 'execute_query' fails due to SQL syntax error, should the agent retry with 'generate_query', or is it unrecoverable?