AI lead-management CRM exposed as an MCP server. Classifies leads using Groq's free-tier Llama model with caching and fallback, manages lead pipeline with status tracking, and provides audit logging.
LeadMind has 8 tools with reasonable naming (verb_noun pattern) and descriptions present for all tools and parameters. However, descriptions are often verbose and include implementation details (e.g., 'Uses a TTL cache first, then Groq...') rather than user-focused intent. Input schemas are present with type declarations, but lack formal constraints (enums for status values, numeric bounds). Output schemas are documented in descriptions but not in structured JSON Schema format. Error handling is minimal, no recovery guidance or actionable error messages visible in code. Tool composition is sound (single responsibility), but some tools could be split (e.g., classify_lead mixes caching logic with classification).
Add a new lead. The message is auto-classified and the lead is saved to the DB with full history. Args: name: Lead's name. contact_info: Email / phone / handle. message: The lead's inquiry message (used for classification). source: Where the lead came from (e.g. 'Website Form', 'Referral'). Returns: {id, name, status, classification}
Bulk import leads from CSV text. Required columns: name, message. Optional: contact_info, source. Uses caching to avoid redundant API calls on duplicate-looking entries. Args: csv_data: CSV text with header row. Example: name,contact_info,message,source John Doe,[email protected],"Interested in pricing",Website Form Returns: {imported, errors, leads: [...]}
Classify a lead message into Hot / Warm / Cold with confidence + reasoning. Uses a TTL cache first, then Groq (Llama-3.3-70b-versatile, free tier), then a rule-based fallback if the API limit is hit — so this tool never fails under public load. Args: text: The lead's message text. Returns: {status, confidence, reasoning, source}
Get the full timeline of status changes and interactions for one lead. Args: id: Lead ID. Returns: {lead: {...}, history: [...]}
Status parameter lacks enum constraint. Description says 'One of Hot | Warm | Cold | Converted | Lost' but no formal enum in schema. LLMs may hallucinate invalid values; validation happens at runtime only.
Descriptions embed implementation details (TTL cache, Groq API, fallback logic) instead of user-focused intent. 'Uses a TTL cache first, then Groq...' is internal; users care about 'Classifies lead sentiment.' This wastes tokens and confuses LLM selection logic.
Output schemas documented only in text descriptions, not in structured JSON Schema. LLMs cannot parse 'Returns: {count, filter, leads: [...]}' reliably. Formal output schema required for downstream tool chaining.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 66 | 2026-07-28+ | v2 |
Get aggregate pipeline stats: total leads, breakdown by status and source, average response time, conversion rate, and Groq usage counts for free-tier monitoring.
Get all leads, optionally filtered by status. Args: status: One of Hot | Warm | Cold | Converted | Lost. Empty string returns all. Returns: {count, filter, leads: [...]}
Generate an AI-recommended next action for a lead based on its history and current status. Falls back to rule-based suggestion if Groq limit hit. Args: id: Lead ID. Returns: {lead_id, lead_name, current_status, suggestion, source}
Manually override a lead's status. Logs the change with timestamp. Args: id: Lead ID. status: One of Hot | Warm | Cold | Converted | Lost. Returns: {id, old_status, new_status, updated_at}
No error handling guidance visible. Code raises exceptions but does not provide recovery hints (e.g., 'Invalid status, valid options are: Hot, Warm, Cold, Converted, Lost'). LLMs cannot self-correct without actionable error messages.
tool_get_lead_stats has empty input schema (no parameters documented). Description does not explain what fields are returned or when to call it. Minimal documentation for a discovery/reporting tool.