Loading ToolBench
MCP Server for Agno framework architecture validation and knowledge RAG.
Server has well-structured tool definitions with complete JSON schemas and descriptions for all 6 tools. Naming follows verb_noun convention (read_, query_, validate_, explain_, get_). Descriptions are substantive (100-300+ chars), exceeding the 10-1024 baseline. However, several parameters lack descriptions entirely, and output schemas are not documented. Error handling guidance is absent. Tool composition is sound, each tool has a single responsibility. No security issues detected (read-only operations, no secrets in params).
Provide detailed explanations and fix patterns for validation issues returned by the architecture validator.
Returns a complete end-to-end architecture plan for building an Agno agent system. Covers the full stack: Agno SDK (agents, tools, memory), AgentOS (FastAPI serving, async runtime), Agno UI (playground, monitoring), MCP layer (IDE integration), and production infrastructure. Call this tool whenever a developer asks how to build, structure, or deploy an agent — before writing any code.
Retrieve relevant Agno framework patterns from curated test-based RAG index, filtered by version.
Deep inspection of a specific local symbol (e.g. a Pydantic model). Returns exact code definition and dependencies.
Lightweight scan of the project structure. Returns file paths and top-level symbols (classes/functions) without internal details to minimize context usage.
Output schemas not documented. Tools return results but LLMs cannot plan downstream calls without knowing response structure (fields, types, pagination). Pattern requires 'Document the output schema.'
Multiple parameters lack descriptions. 'include_definition' (query_local_architecture), 'range' object properties (validate_architecture_basics), 'issues' array items (explain_validation_failure) have no guidance. Pattern requires 'Every parameter needs a description.'
No error handling guidance. Tools do not document what errors can occur, whether they are retryable, or what the LLM should do next. Pattern requires 'Error responses must tell the LLM what to do next.'
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 61 | 2026-07-28+ | v2 |
Validate Python code against core architectural axioms (Pydantic usage, async compliance, no global state). Supports raw code or file-based validation.
validate_architecture_basics uses oneOf constraint but does not explain mutual exclusivity in parameter descriptions. LLMs may pass both 'code' and 'filepath' unless explicitly told not to. Pattern requires 'If parameters are mutually exclusive, state this in descriptions.'
query_framework_knowledge returns 'top_k' results but no pagination or total count documented. If RAG index is large, LLMs cannot iterate through all results. Pattern requires 'Tools returning lists should accept page/offset and limit parameters and return a total count.'