Loading ToolBench
Assert that an MCP server's configuration profile exposes exactly the tools it should, and none it shouldn't — a permission/capability boundary tester, not a schema-drift tool.
mcp-capguard is a testing/validation tool, not a production MCP server. It exposes 3 tools with minimal descriptions (10-30 chars, well below the 50-200 char baseline for LLM-optimized descriptions). Tool schemas are present but sparse: 'search' and 'get_document' have single string parameters with minimal context; 'delete_document' lacks any indication it's destructive despite the Risk annotation. No output schemas documented. No error handling guidance. The tool definitions appear inferred from example code rather than explicitly registered with full metadata. This is a utility library, not a production-grade agent tool.
Permanently delete a document.
Fetch a document by id.
Search documents.
Descriptions critically short (10-30 chars). Baseline for LLM-optimized descriptions is 50-200 chars. 'Search documents' and 'Fetch a document by id' lack context on when to use, what to expect, or dependencies.
No output schemas documented. LLMs cannot plan downstream calls or extract structured data without knowing what fields to expect.
delete_document marked DESTRUCTIVE but description does not state it modifies state or is irreversible. Agents need explicit warnings on destructive operations.
No error handling guidance. Tools return plain strings with no structure for success/failure, invalid input, or recovery steps.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 37 | 2026-07-28+ | v2 |
Parameter 'query' and 'doc_id' lack format constraints, ranges, or validation rules. LLMs cannot infer valid input formats.