Loading ToolBench
An MCP server for interacting with Outline's API
The Outline MCP server demonstrates solid definition quality overall. Tool naming is consistently verb-based (archive_, ask_, create_, delete_) and follows the required pattern. Descriptions are present on all 8 tools and range from adequate to comprehensive. Input schemas are properly defined using Zod with type information and descriptions for all parameters. However, there are notable gaps: output schemas are not documented anywhere in the codebase; error handling lacks recovery guidance in descriptions; some parameter descriptions could be more actionable; and no tool annotations (readOnlyHint/destructiveHint) are visible despite the clear risk levels indicated.
Archive a document
Query documents with natural language. This method allows asking direct questions of your documents - where possible an answer will be provided. Search results will be restricted to those accessible by the current access token. Note that "AI answers" must be enabled for the workspace.
Create a new collection
Create a new comment on a document
Create a new document
Create a template from an existing document. This method allows you to create a new template using an existing document as the basis.
No output schemas documented. Tools lack return type definitions, preventing LLMs from understanding what fields to expect in responses. This violates the pattern:tool requirement and forces LLMs to parse responses blindly.
No tool annotations visible. Despite clear risk levels (WRITE, DESTRUCTIVE, READ_ONLY), tools lack destructiveHint/readOnlyHint annotations that would inform LLM planning and safety reasoning.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | B | 72 | 2026-07-28+ | v2 |
| 2026-03-09 | B | 73 | - | v1 |
Delete a comment from a document
Delete a document
Destructive operations lack confirmation/dry-run guidance. delete_comment and delete_document descriptions do not mention confirmation requirements or recovery paths, violating pattern:confirmation-request.
Error handling not addressed in tool descriptions. No guidance on what to do if a document ID is invalid, a collection doesn't exist, or the workspace lacks AI-answer capability (required for ask_documents). Violates pattern:recovery-guide.
Several parameter descriptions are generic or lack actionable constraints. 'Permission level for the collection' does not explain when to use read vs read_write. 'Content of the comment in markdown format' does not specify length limits or content restrictions.
ask_documents description mentions 'AI answers must be enabled' but offers no guidance on how to detect or handle this prerequisite. If the feature is disabled, the tool will fail without recovery path.
No pagination guidance documented. ask_documents may return large result sets but does not mention limit/offset parameters or how many results to expect. Pattern:paginated-result not fully implemented.
Tool dependencies not documented. create_comment requires a valid documentId; create_document requires a valid collectionId. No discovery guidance (e.g., 'use create_collection first if unsure').