Loading ToolBench
MCP server providing access to Agentcy marketing data query, web research, and data sources discovery APIs
This server exposes 3 tools via HTTP to an external Agentcy API. All tools have descriptions (good), but parameter schemas are incomplete and lack type information. Tool names follow verb_noun convention (agentcy_query, agentcy_research, agentcy_sources), but descriptions are generic and do not explain WHEN to use each tool or what distinguishes them. Parameters like 'question', 'domain', 'start_date' lack type declarations in the visible schema. No output schema is documented. Error handling in the shell scripts returns HTTP status codes and error messages, but does not guide the LLM on recovery actions. The server acts as a thin wrapper around external APIs, tool definitions should be richer to help LLMs reason about tool selection and composition.
Queries marketing data for a specific client domain via the Agentcy REST API. Returns synthesized insights, not raw data.
Performs web research and competitive intelligence via the Agentcy REST API. Returns synthesized findings with sources.
Lists configured domains and connected marketing data sources.
Input schemas lack type declarations. Parameters 'question', 'domain', 'start_date', 'end_date', 'source_hints' are described as strings but no JSON Schema type field is visible.
No output schema documented. Tools return synthesized text + optional metadata (services/sources), but LLMs cannot plan downstream calls or extract structured data without knowing the response shape. Rubric baseline: '100% of A+ tools have documented return types.'
Tool descriptions do not explain WHEN to use each tool or how they differ. 'Queries marketing data' vs 'Performs web research' is vague, when should an LLM pick agentcy_query over agentcy_research? Descriptions should state prerequisites, use cases, and data freshness. Rubric: 'A tool description must answer: What does it do? When should the LLM call it instead of a similar tool?'
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 48 | 2026-07-28+ | v2 |
Error handling returns HTTP status codes and error messages but does not guide recovery. Shell scripts echo 'Error (400): ...', the LLM cannot infer whether to retry, ask the user, or try a different tool. Rubric: 'Error responses must tell the LLM what to do next.'
Parameter 'source_hints' in agentcy_query is optional and comma-separated, but no enum or format constraint is documented. LLMs may pass invalid source names. Rubric: 'When a parameter accepts one of a known set of values, declare it as an enum.'