Research MCP has 8 tools with reasonable naming and descriptions, but falls short on parameter documentation rigor. All tools are READ_ONLY (good security posture). Tool descriptions are present and contextually helpful (120-180 chars average), meeting the baseline. However, parameter schemas lack depth: many parameters have minimal descriptions, no format constraints, no enums for constrained choices, and no validation guidance. The 'ask-perplexity' tool asks for a 'query' string with no format bounds; 'deepResearch' boolean is under-explained. GitHub search tools mention '2-3 core keywords max' in descriptions (good UX guidance) but lack enum constraints for the 'language' and 'license' parameters. OpenAlex search tools have sparse parameter documentation, 'filter' and 'sort' accept free-form strings with no examples or format guidance. Output schemas are not documented in the tool definitions, forcing LLMs to infer result structure. Error handling is not visible in tool descriptions. Overall, this is a competent read-only research toolkit but lacks the parameter rigor and output documentation expected of production-grade agent tools.

• Add output schema documentation to every tool. Example: 'Returns an object with fields: {results: Array<{id: string, title: string, url: string, score: number}>, total_count: number, next_cursor?: string}'.
• Convert unconstrained string parameters to enums. E.g., github-search-repositories 'language' should become an enum of popular languages (python, javascript, java, go, rust, etc.) with a fallback 'other' option.
• Expand parameter descriptions with examples and syntax. E.g., 'filter: Filter by field values using OpenAlex syntax, e.g., "publication_year:2023", "author.id:A123456". Leave empty for no filter.'
• Document pagination behavior. Add to tool descriptions: 'Results are paginated. Set limit to control page size (1-200, default 10). Use offset (for OpenAlex) or cursor/first parameters (for GitHub) to fetch additional pages.'
• Add error handling guidance to tool descriptions. Example: 'On API error or rate limiting, the tool returns an error message. If rate-limited, wait and retry after 60 seconds. If the query returns no results, try simplifying keywords or adjusting filters.'
• For get-markdown, clarify when to use it: 'Use this tool to extract structured content from technical documentation or blog posts. Returns the full page as markdown. Best for single-page docs; avoid for JavaScript-heavy sites or infinite-scroll pages.'
• Specify default values for optional parameters in descriptions. E.g., 'first: number of results to return (default 10, max 100)' not just '(default 10, max 100)'.