The server defines 4 tools with visible schemas and descriptions, but exhibits significant quality gaps. All tools have input schemas with typed parameters and descriptions present, meeting baseline structure. However, descriptions are inconsistent in clarity and actionability. Tool naming follows verb_noun convention (list_, get_, extract_, analyze_) which is acceptable. Critical issues: (1) Output schemas are NOT documented anywhere, the code returns TextContent with JSON strings, but the LLM cannot know what fields to expect in the parsed response. (2) No error handling guidance is visible; error returns are bare strings like 'Error: Session not found' without recovery suggestions. (3) Session-based architecture (session_id parameter in every tool) is awkward, LLMs must track session IDs manually; this design does not match the chat data model where users expect stateless, name-based queries. (4) Descriptions are generic/weak for distinguish ability: list_flows vs get_flow_details both describe retrieving HTTP data, creating ambiguity about when to call which. (5) No pagination or result limits documented despite tools that could return large result sets. (6) Parameter descriptions lack actionable constraints (e.g., 'session_id' is just described as 'The ID of the session', no guidance on format, how to obtain one, or error handling if invalid).

• Add output schema documentation to every tool. Specify the structure LLMs receive: 'Returns: { flows: [{ index: int, method: string, url: string, status: int }] }' with all fields, types, and cardinality.
• Disambiguate list_flows vs get_flow_details. Clarify: 'list_flows shows all flows in a session (summary only); get_flow_details retrieves full request/response bodies for specified flows. Use list_flows for discovery, then get_flow_details to inspect specific ones.'
• Add error recovery guidance. Replace 'Error: Session not found' with 'Session ABC123 not found. Valid sessions: [list]. Call list_sessions() to discover available sessions or create a new one.'
• Document session_id format and discovery. Add: 'session_id: string (UUID format, obtained from mitmproxy dump filename or list_sessions call)' and provide or reference a tool to list/create sessions.
• Implement pagination for list_flows and get_flow_details. Add 'limit' (default 20, max 100) and 'offset' parameters. Return total_count so LLM knows when to fetch next page.
• Add per-parameter format/constraint hints. 'flow_indexes: array of integers (0-based, e.g., [0, 5, 10])' and 'json_paths: JSONPath expressions (RFC 9535, e.g., $.data.users[0].email)'.
• Document what extract_json_fields returns when paths do not exist. Specify: 'Returns object mapping each path to matched value or null if not found. Example: {"$.data.users": [...], "$.missing.path": null}'.
• Document analyze_protection output structure. Specify fields like protection_type (string enum: cloudflare, recaptcha, etc.), challenge_type, required_headers, scripts_found (count), and recommended_actions (array).