AnalyticDB PostgreSQL MCP Server serves as a universal interface between AI Agents and AnalyticDB PostgreSQL databases. It enables seamless communication between AI Agents and AnalyticDB PostgreSQL, helping AI Agents retrieve database metadata and execute SQL operations.
This server has 15 tools with mixed definition quality. Naming is action-oriented (execute_, adbpg_, call_) but some tool names are overly compound (e.g., 'adbpg_graphrag_upload_decision_tree'). Descriptions are present but brief (avg ~60 chars), often lacking context about when to use the tool or prerequisites. Schemas are visible for all tools and include required fields, but most parameters lack detailed type constraints (enums, ranges, patterns). Error handling is minimal, no guidance on recovery or categorization. Security concerns exist: raw SQL parameters invite injection; no validation shown. Output schemas are not documented. Overall falls into the 'fair' (C) range due to significant gaps in parameter descriptions, missing output documentation, and incomplete error handling.
Append a subtree to an existing decision tree at the node specified by root_node_id.
Delete a sub-decision tree under the node specified by root_node_entity.
Execute graphrag query operation
Reset the decision tree in the tree query mode
Execute graphrag upload operation
Upload a decision tree with the specified root_node. If the root_node does not exist, a new decision tree will be created.
SQL injection vulnerability in execute_select_sql, execute_dml_sql, execute_ddl_sql: raw SQL queries accepted as strings with no validation or parameterization guidance. LLMs can be tricked into passing malicious SQL.
Destructive operations (execute_ddl_sql, adbpg_graphrag_delete_decision_tree, adbpg_llm_memory_delete_all) lack confirmation/dry-run mechanism. Agents can irreversibly delete data without safeguard.
Most tool descriptions are under 100 characters and lack context: when to use the tool, what it returns, prerequisites, or dependencies. Descriptions do not answer 'What does it do? When should I call it? What do I get back?' Required for LLM tool selection.
Inferred effective spec: 2026-07-28+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | C | 62 | 2026-07-28+ | v2 |
| 2026-03-09 | D | 53 | - | v1 |
Execute llm_memory add operation
Execute llm_memory delete_all operation
Execute llm_memory get_all operation
Execute llm_memory search operation
Execute ANALYZE command to collect table statistics.
Execute (CREATE, ALTER, DROP) SQL statements to manage database objects.
Execute (INSERT, UPDATE, DELETE) SQL to modify data in ADBPG database.
Execute SELECT SQL to query data from ADBPG database. Returns data in JSON format.
Get query execution plan.
No output schemas documented for any tool. LLMs cannot anticipate what fields to expect in responses, forcing downstream reasoning about field names and types. Increases hallucination and wasted tokens on guessing.
No input validation or error handling visible. Tools do not guide LLMs on recovery. Example: if execute_select_sql gets invalid SQL, the error should suggest checking syntax or consulting EXPLAIN, not return a raw database error.
Parameter descriptions contain typos: 'root_noot' instead of 'root_node' in adbpg_graphrag_upload_decision_tree, adbpg_graphrag_append_decision_tree, adbpg_graphrag_delete_decision_tree. Confuses LLM parameter mapping.
Enum constraints not formalized in JSON Schema. Example: adbpg_graphrag_query's query_mode lists options inline in description ('bypass,naive,local,global,hybrid,mix,tree') rather than as formal enum in schema. LLMs cannot reliably parse inline text enums.
No pagination or result size limits documented. Tools like adbpg_llm_memory_get_all, adbpg_llm_memory_search could return unbounded results, blowing context window. No page/offset/limit parameters visible.
Overly compound tool names: adbpg_graphrag_upload_decision_tree, adbpg_graphrag_append_decision_tree. These 40+ character names are harder for LLMs to disambiguate; 'upload_decision_tree' + 'append_decision_tree' pattern signals multiple responsibilities (upload + decide, append + decide). Consider 'graphrag_upload' and 'graphrag_append' with context as separate params.