Loading ToolBench
Read-only MCP server for consulting a project's wiki. Serves wiki pages, sources, and index structure from an open-wiki project.
Four read-only tools with basic descriptions and empty input schemas. Tool names lack action verbs (ow_index, ow_read_page, ow_sources, ow_read_source are prefixed with 'ow_' but lack clear verb-noun structure). Descriptions are present but generic (34 - 194 chars, within baseline range). Critical gap: ow_read_page and ow_read_source have input parameters (slug, id) with type definitions visible in the rubric summary, but no descriptions for those parameters are evident in the source excerpt. No output schemas documented. No error handling guidance. No tool annotations. Composition is reasonable (each tool does one thing), but parameter documentation is incomplete.
List every entity page in the wiki as structure: slug, title, type, status, and whether the index links to it (false marks an orphan).
Return a page whole: its full markdown and its parsed frontmatter.
Return a source's text.md when it has one. A source with none reports what it is instead — its title, kind, original filename, status and description — because this project stores sources and does not read them (adr:0021). The original file is not reachable from here.
List every source under raw/: its frozen id, its manifest (title, kind, original), and whether text.md is present.
Tool names lack action verbs. 'ow_index', 'ow_read_page', 'ow_sources', 'ow_read_source' are prefixed with domain ('ow_') but do not follow verb_noun convention (e.g., 'list_pages', 'get_page', 'list_sources', 'get_source'). LLMs rely on verb-first naming to infer intent.
Input parameters lack descriptions. ow_read_page accepts 'slug' and ow_read_source accepts 'id', but no parameter descriptions are visible in the source. LLMs cannot infer what 'slug' or 'id' format is expected (e.g., 'wiki/my-page' vs 'my-page', frozen ID format).
No output schemas documented. Tool descriptions state what is returned (e.g., 'full markdown and parsed frontmatter', 'text.md or metadata'), but no structured schema is provided. LLMs cannot plan downstream operations or extract specific fields.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 48 | 2026-07-28+ | v2 |
No error handling guidance. Tools are read-only, but no error responses document what happens if a slug/id is not found, or if the wiki is malformed. LLMs have no recovery path.
No tool annotations. Tools lack readOnlyHint, destructiveHint, or idempotentHint. Although all four are read-only, explicit annotations would clarify this to the agent and enable safer planning.