Loading ToolBench
Trusted document supply for AI coding agents: a read-only MCP server that enforces document governance by verifying documents against a trust table and dependency fingerprints before serving them to agents.
Three well-named, read-only tools with clear governance semantics. Tool names follow verb_noun pattern (get_document, list_documents, document_dependencies). Descriptions are substantive (150 - 250 chars) and explain WHAT, WHEN, and WHY. Input schemas are present with type declarations and descriptions. However, output schemas are not documented in the visible code, LLMs cannot predict response structure. Error handling is mentioned in descriptions (e.g., 'Refuses with a structured error code') but recovery guidance is sparse. No parameter enums, no pagination support, no per-item success/failure for batch operations.
Retrieve the declared dependencies of a governed document and verify they match the current working tree. Returns the list of source files the document depends on and whether the dependency fingerprint still matches the recorded value.
Retrieve a governed Markdown document from the repository. Returns the document content only if it passes trust verification: the document exists, is readable, matches its recorded content hash, and all declared dependencies remain unchanged. Refuses with a structured error code and source locations to read instead if verification fails.
List all governed documents in the repository with their current verification status. Returns document paths, types, and usability status. A document listed as usable may still be refused by get_document if its dependencies have changed since the list was generated.
Output schemas not documented. LLMs cannot predict response structure (fields, types, pagination). Descriptions mention 'returns' but do not specify JSON schema.
list_documents accepts optional document_type parameter but does not declare it as an enum. Free-form string invites hallucinated values like 'legal' or 'spec' instead of the five valid types.
No pagination support. list_documents returns all governed documents without limit or offset. Large repositories risk context window exhaustion.
Error recovery guidance is implicit. Descriptions mention 'structured error code' and 'source locations to read instead' but do not enumerate error codes or explain when to retry vs. escalate.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 68 | 2026-07-28+ | v2 |
Path parameter descriptions do not specify validation rules. No mention of path traversal guards, allowed characters, or max length. LLMs may pass '../../../etc/passwd' or other malicious paths.