Loading ToolBench
MCP server for filtering vector database metadata using query and filter expressions
This server exhibits critical quality failures across nearly every dimension. The single tool 'execute_skill_action' has no visible input schema, no parameter descriptions, and a bare-minimum description. The code provided does not constitute a functional MCP server implementation, it is a stub that prints JSON and exits. No actual tool invocation, parameter handling, or MCP protocol compliance is evident. The client.py and example_usage.py modules demonstrate vector DB filtering logic but are never integrated into the MCP server. The implementation cannot accept tool calls, validate parameters, or return structured results.
Execute skill action for vector database metadata filtering
No input schema visible or defined for execute_skill_action. The tool accepts parameters but provides no JSON Schema definition, type information, or constraints.
Tool description is extremely sparse ('Execute skill action for vector database metadata filtering', 66 characters). While above the 20-char threshold, it lacks WHAT the tool does operationally, WHEN to use it, what parameters it expects, and what it returns. LLMs cannot infer these details.
No parameter descriptions. The tool likely accepts parameters (query, filters, or similar) based on the client.py logic, but the MCP server stub provides no documentation of what parameters exist, their types, constraints, or meanings.
No output schema documented. The tool returns some result (likely a filtered record set or applied filter expression) but the MCP server does not specify what fields, types, or structure the response contains.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 12 | 2026-07-28+ | v2 |
Tool name 'execute_skill_action' is generic and vague. It does not follow the verb_noun pattern or clearly signal what action occurs. 'filter_vector_db_metadata', 'search_with_metadata_filter', or 'apply_metadata_filter' would be far more specific and action-oriented.
MCP server implementation is non-functional. The run_mcp_server() function prints a static JSON status and exits. It does not implement the MCP protocol, no request/response handling, no tool invocation dispatch, no error handling. This is a skeleton, not a working server.
Client logic (AgentVectorDBMetadataFilterClient.build_filter) is disconnected from the MCP server. The server does not invoke this client, accept tool calls, or integrate the filtering logic into the MCP protocol flow.
No error handling, validation, or error guidance. The server provides no mechanism to validate inputs, catch failures, or guide LLM recovery. A malformed call returns nothing actionable.