Loading ToolBench
MCP server providing workspace knowledge tools for document search, retrieval, and analysis with evidence-based citation and tool-loop support
KnowVault demonstrates solid definition quality with complete input schemas, clear descriptions emphasizing pagination and scope limitations, and consistent verb-noun naming (search_, grep, list_, read, related, sources, refresh). All 7 tools have descriptions (avg ~150 chars) and documented parameters with types. However, output schemas are not visible in the provided source, parameter descriptions lack format constraints and validation rules, and error handling guidance is absent. The descriptions are notably verbose about pagination caveats but lack actionable recovery guidance for failure cases. No tool annotations (readOnlyHint/destructiveHint) despite clear risk classifications.
Search workspace evidence by regex pattern. Matches select evidence to read; the first page does not establish completeness. For broad questions, examine relevant sections and continuations, and disclose the scope actually checked.
List workspace document and evidence object inventory. This is document metadata for navigation, not evidence of its subject matter. For an overview, move from the first small inventory page to reading different relevant documents and state the scope actually checked. An explicitly requested exhaustive list or total still requires checking the full relevant scope.
Read the relevant content before forming claims. Continue when needed for the requested scope; later automatic citation validation cannot assess omitted context.
Trigger or check the status of workspace data source synchronization.
Find evidence related to a given fragment or address. Returns related items for navigation and context.
Search workspace evidence by query. Matches select evidence to read; the first page does not establish completeness. For broad questions, examine relevant sections and continuations, and disclose the scope actually checked.
Output schemas not documented. LLMs cannot plan downstream tool calls or extract required fields (e.g., what does knowvault_search return? Does it include fragment_id for chaining to knowvault_read?). Breaks tool composition.
No error handling guidance. Descriptions lack recovery hints (e.g., 'If workspace_id is invalid, call knowvault_sources() to list available workspaces'). Agents cannot self-correct on failures.
No tool annotations. knowvault_refresh is marked WRITE but lacks destructiveHint annotation. knowvault_search/grep/list_objects/read/related/sources lack readOnlyHint. Agents cannot infer safety/idempotency without explicit hints.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 57 | 2025-06-18+ | v2 |
List workspace data sources and their synchronization state. This reports connection and synchronization state, not the knowledge contained in documents. Use these technical fields in an answer only when the user asks about source or connection state.
Parameter descriptions lack format constraints. 'pattern' in knowvault_grep has no regex syntax guidance. 'limit' parameters lack min/max bounds (e.g., 1 - 100). 'offset' lacks explanation of pagination semantics. Invites invalid LLM input.
workspace_id is optional in most tools but required for all operations. Descriptions should clarify: is there a default workspace? Or must agents always provide it? Ambiguity forces extra discovery calls.