Loading ToolBench
Native MCP inspector: connect, browse, call, replay, and diff MCP servers. Also includes mcpi-cli for CI snapshots and contract validation.
This is a meta-MCP server (router/inspector) with 6 tools. Critical issues: (1) Tool descriptions are extremely brief (5-30 chars), well below the 10-1024 char baseline and failing the 20-char minimum for description scoring. (2) Input schemas are present but minimal, most tools have only 1-2 parameters with basic type info; no enums, ranges, or validation constraints visible. (3) Tool naming is reasonable (verb-noun style: find_tools, call_tool, search, summarize) but lacks clarity on when to use each. (4) No output schemas documented anywhere in the source. (5) Error handling and recovery guidance absent. (6) The 'deprecated_tool' with empty schema is a red flag. (7) Tool composition is weak, find_tools and call_tool are generic routers that don't guide the LLM on which upstream server to target.
Routes tool calls across upstream MCP servers
Scheduled for removal.
Echo the message back.
Routes find_tools across upstream MCP servers with a TypeSafe Jev decision
Search the index.
Summarise a document.
Tool descriptions critically short (5-30 chars). Baseline is 10-1024 chars; descriptions under 20 chars score 0-20. Examples: 'Routes find_tools across upstream MCP servers with a TypeSafe Jev decision' (73 chars, but vague on 'TypeSafe Jev'); 'Echo the message back.' (22 chars, minimal). LLMs cannot determine when to select these tools.
No output schemas documented. Tools return results but LLMs have no visibility into response structure. Baseline: 100% of A+ tools document return types. This prevents agents from planning downstream calls and extracting required fields.
deprecated_tool has empty schema ({}) and no description. This tool should be removed or properly defined.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 43 | 2026-07-28+ | v2 |
Input schemas lack constraints. 'search' has limit (1-100) and mode enum, which is good, but 'find_tools' and 'call_tool' accept free-form strings with no validation hints. No regex patterns, length limits, or format guidance in descriptions. Baseline: 94% of A+ tools constrain inputs.
No error handling or recovery guidance. Tools provide no actionable error messages. If 'call_tool' fails, the LLM has no guidance on retry, fallback, or next steps. Baseline: A+ tools include recovery hints.