MCP server for searching and retrieving content from the Atlassian Community forums, including Q&A posts, articles, and user-generated content
Server has 7 read-only tools with consistent naming (verb_noun pattern: search*, get*) and complete Zod schemas. All tools have descriptions (10-80 chars) and input parameters with types. However, descriptions are terse and lack LLM-optimized guidance on WHEN to use each tool vs. alternatives. Output schemas are not documented, responses are JSON-stringified text, not structured objects. Error handling returns JSON-wrapped errors but lacks recovery guidance. No tool annotations (readOnlyHint, etc.). Pagination is well-implemented across all tools (limit 1-100, offset, sortOrder). Missing: output field documentation, dependency hints between tools, and actionable error messages.
Get most recent posts (look for new posts across all tags)
Get most recent posts for a tag (look for new post in one or more tags)
Get answers for a post (get all comments with a parent post being a specific id)
Get top posts by views for a tag (top posts by page views for a given tag)
Get content by user (get details on users who post, including other posts they've made)
Search community posts by query and tags (look for content in specific tags)
Search community posts by query (look for content in all tags)
Output schemas not documented. Tools return JSON-stringified text in a generic 'content' wrapper. LLMs cannot infer response structure (which fields exist, types, required vs. optional). This forces parsing unstructured output and wastes tokens.
Descriptions are terse (10-80 chars) and lack LLM-optimized guidance. E.g., 'Search community posts by query (look for content in all tags)' does not explain WHEN to use searchCommunity vs. searchByTags, or what the response contains. Baseline for A+ tools is 50-200 chars with clear WHAT/WHEN/WHY.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 51 | <=2025-11-25 | v2 |
Error handling returns JSON-wrapped error messages but lacks recovery guidance. E.g., 'error: User not found' does not tell the LLM what to do next (retry? call search_users? ask user?). Errors should be categorized as retryable, user-fixable, or fatal.
No tool annotations. Tools lack readOnlyHint, destructiveHint, or idempotentHint metadata. All 7 tools are read-only and idempotent, annotating them enables agents to reason about safety and retry behavior without reading descriptions.
No dependency hints between tools. LLMs cannot infer that searchCommunity and searchByTags are alternatives, or that getPostAnswers requires a postId from a prior search. Descriptions should guide multi-step planning.