Advanced research engine with live parameter controls for web scraping, content gathering, analysis, and asset generation across multiple platforms
The server has 9 tools with reasonable naming (verb_noun style) and present descriptions. However, there are significant gaps: (1) descriptions are generic and lack WHEN/WHY guidance; (2) output schemas are completely undocumented, no tool states what fields it returns or their types; (3) error handling is minimal, most tools return JSON status strings without actionable recovery guidance; (4) parameters lack constraints (enums for source_type, output_format, content_type, tone; ranges for depth, max_results, relevance_threshold); (5) the server is stateful (global engine, current_task variables) rather than stateless, forcing multi-call workflows instead of self-contained requests. Tool names are clear (start_research, get_research_status, pause_research, resume_research, stop_research, update_parameters, get_results, export_results, build_content) but the definitions lack production rigor.
Build structured content from research results
Export research results to a file
Get the current status of the research task
Get the current research results
Pause the current research task
Resume the paused research task
Start a research task with specified parameters
No output schemas documented. Tools return JSON strings (e.g., get_research_status returns {is_running, is_paused, progress, results_count, stats}) but the response structure is not formally declared. LLMs cannot reliably extract fields or plan downstream calls without documented output schemas.
Parameters lack enum constraints. source_type accepts 'all, academic, news, technical, social' as free-form string with no enum declaration. output_format, summary_length, content_type, and tone suffer the same issue. Free-form strings invite hallucinated values (e.g., 'academic_journals', 'professional_friendly'). Replace with JSON Schema enums.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | C | 61 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 17 | - | v1 |
Stop the current research task
Update research parameters while task is running
Stateful global engine (engine, current_task variables). The server maintains state across requests, breaking stateless MCP design. start_research checks 'if engine.is_running' and then modifies global state. This forces callers to sequence requests (check status, then start, then poll) instead of each call being self-contained. Refactor to accept a session_id or research_id parameter and store state server-side (e.g., in a database) keyed per request.
Error handling lacks recovery guidance. Tools return JSON with {status: 'error', message: '...'} but do not guide the agent on next steps. E.g., start_research returns 'A research task is already running. Please stop it first.', no mention of calling stop_research() by name. Error responses should be actionable: 'Research task already running. Call stop_research() first, then retry.'
Missing per-parameter constraints. depth (1-10), max_results (default 20), relevance_threshold (0.0-1.0), detail_level (1-10) lack explicit min/max in schema. LLMs can pass depth=0 or depth=100, breaking logic. Add JSON Schema minimum/maximum properties.
Tool descriptions are generic and lack WHEN/WHY guidance. E.g., 'Get the current status of the research task' (39 chars) does not explain what fields are returned, when to call it, or how it relates to start_research. Production descriptions (50-200 chars) should state: 'Returns progress (0-100), result count, and running/paused state. Call this to check if your research has completed before calling get_results().'
No documentation of irreversible operations. start_research, export_results, and build_content trigger long-running actions but descriptions do not warn agents these calls have side effects. Agents should know: is this retriable? Does it create a new record each time? Is it idempotent?
Missing tool annotations. No destructiveHint, readOnlyHint, or idempotentHint declared. For MCP spec compliance (2026-07-28), tools should declare their side effects. start_research and export_results are destructive; get_research_status and get_results are read-only. These annotations help agents reason about safety and retryability.