Loading ToolBench
List, create and deploy your Prefab project templates from any MCP agent.
Prefab MCP has solid naming conventions (verb_noun pattern: list_, describe_, preview_, deploy_) and comprehensive descriptions (avg 180 chars, well above 10-char minimum). All 6 tools are explicitly registered with schemas and annotations. However, parameter descriptions are sparse, most lack detail about format, constraints, or dependencies. The 'values' parameter across preview_deploy and deploy_template is documented only as 'Placeholder values, keyed by placeholder key, e.g. {"client_name": "Acme"}' without explaining valid keys, required vs optional, or format constraints. Error handling is present but generic ('No template matching...'). Output schemas are not formally documented, responses are text-wrapped without structured field definitions. Tool annotations (readOnlyHint) are correctly applied to read operations.
Create a template's folders and files inside a destination folder, filling in the values given. This writes to disk. Call preview_deploy first to see what it will make and whether Prefab can write there.
The full structure of one template: every folder and file it creates, the placeholders it asks for, the actions attached to it, and any templates it embeds.
Library-wide placeholders, actions and saved value presets - the things defined once and available to every template.
Every template in the user's Prefab library: what it is called, what it creates, and the placeholder values it needs at deploy time.
Folders Prefab is watching, and the rules that run on files landing in them. Explains why a file on this Mac was renamed, converted or moved.
What deploying a template into a folder would create, with the given values filled in. Creates nothing. Reports any required value still missing, and whether Prefab has write access to the destination. Worth calling before deploy_template.
Parameter 'values' lacks constraint documentation. No enum of valid keys, no indication of required vs optional fields, no format specification. LLMs cannot infer which placeholders are mandatory or what format they expect.
Output schemas are not formally documented. Tools return text-wrapped JSON without declaring field types, required fields, or structure. LLMs cannot reliably extract or chain results.
Error messages are generic and do not guide recovery. 'No template matching...' tells the LLM to call list_templates, but other failure modes (permission denied, invalid destination path, missing required values) lack actionable guidance.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 60 | 2026-07-28+ | v2 |
No pagination or result limits documented. list_templates, list_watched_folders, and list_globals may return unbounded results. No mention of max items, pagination tokens, or truncation behavior.
deploy_template is destructive (writes to disk) but lacks confirmation or dry-run capability. No isError flag in responses to distinguish success from partial failure. Agents cannot safely preview consequences before committing.