The 'scan' tool has a reasonable name (verb-noun) and acceptable description (89 chars), but critical gaps undermine quality. The input schema is present with 4 parameters, each with type and description, meeting baseline requirements. However, the schema lacks critical constraints: 'window_sec' and 'min_cluster_accounts'/'min_burst_accounts' are unbounded integers with no min/max validation guidance. The 'posts' parameter description is vague ('JSON array or JSONL of post objects'), it does not specify required fields, format strictness, or error handling for malformed input. No output schema is documented anywhere in the visible code (electionlens/mcp_server.py uses a wrapper 'build_mcp_server' that abstracts the actual schema). The tool description does not explain what 'coordinated inauthentic behavior' means operationally, when to call it vs. alternatives, or what the output structure contains. Error handling is absent, no guidance on what happens if posts are malformed, if thresholds are unreasonable, or how to interpret results. Since the visible schema is incomplete and output is undocumented, the effective score is lower.

• Inline the tool definition directly in electionlens/mcp_server.py instead of delegating to cognis_core.mcp.build_mcp_server. Make schema, output structure, and error handling visible and auditable.
• Document the output schema explicitly. Example: 'Returns {coordination_index: float, risk_level: string, copypasta_clusters: array, burst_windows: array, narrative_spikes: array, top_accounts: array}'. Specify each field's type and meaning.
• Add min/max constraints to numeric parameters: window_sec (1 - 3600), min_cluster_accounts (2 - 100), min_burst_accounts (2 - 100). Include these in the parameter description.
• Clarify the 'posts' parameter: specify required fields (id, account, timestamp, text), acceptable formats (JSON array or JSONL), size limits, and timestamp format (ISO 8601). Example: 'Array of post objects, each with: id (string), account (string), timestamp (ISO 8601), text (string), hashtags (optional array of strings). Max 10,000 posts per call.'
• Add error handling guidance: 'If posts are malformed, returns error with details. If thresholds are unreasonable (e.g., min_cluster_accounts > distinct accounts), returns empty results with explanation. Call with default thresholds (window_sec=120, min_cluster_accounts=3, min_burst_accounts=4) for typical election monitoring.'