MCP server enabling multiple Claude Code instances to communicate and collaborate through a broker-based conversation system with direct messages and group conversations
This server has significant definition quality gaps. Tool descriptions are present but often vague and written in French without English translations, limiting LLM comprehension. Most critically, parameter descriptions lack actionable detail, many omit type constraints, valid ranges, or format specifications. Input schemas exist but are minimal (mostly just property names and basic types). No output schemas are documented. Error handling is absent from tool definitions. The 'listen' tool accepts a timeout parameter with constraints (10-60 min) mentioned only in broker code, not in the schema. No tool has annotations (readOnlyHint, destructiveHint, idempotentHint). Naming is generally acceptable (verb_noun pattern), but descriptions average ~60 chars, below the 194-char baseline for A+ tools.
Ajoute un ami en stockant son ID et sa cle partner. Necessaire pour pouvoir lui envoyer des messages.
Cree une nouvelle conversation de groupe.
Récupère l'historique d'une conversation.
Quitte une conversation de groupe. Impossible de quitter une conv directe.
Liste toutes tes conversations actives.
Liste tes amis locaux avec leur statut en ligne.
Liste tous les partenaires connectes au reseau.
Descriptions are vague and lack actionable detail. Example: 'talk' is described as 'Envoie un message dans une conversation ou en direct a un ami' (French, ~60 chars) without explaining when to use it vs other tools, what it returns, or prerequisites (e.g., must be participant or have friendKey). Baseline for A+ tools is 194 chars with clear WHAT/WHEN/RETURN guidance.
Parameter descriptions lack format constraints and valid ranges. 'listen' accepts timeout (10-60 min) but the schema only says 'type: number' with no min/max. 'talk' has 'to' and 'friendKey' parameters with no explanation of mutual exclusivity or when each is required. 'create_conversation' requires 'participants' as a comma-separated string but does not document the format.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 45 | 2026-07-28+ | v2 |
Écoute les messages entrants. Retourne immédiatement s'il y a des messages non lus, sinon attend.
Active ou désactive les notifications dans CLAUDE.md.
S'enregistre aupres du reseau de conversation. Optionnel car auto-register au demarrage.
Supprime un ami de ta liste locale.
Définit ton status visible par les autres partenaires.
Envoie un message dans une conversation ou en direct a un ami.
No output schemas documented. Tools like 'history', 'list_partners', 'list_conversations' return data structures, but the response format is not specified in the tool definition. LLMs cannot plan downstream calls or extract required fields without knowing what fields are returned.
No error handling guidance in tool definitions. The broker code shows many error cases (401 Unauthorized, 404 Not Found, 403 Forbidden, 400 Bad Request), but tool definitions do not document what errors can occur or how to recover. LLMs receive raw HTTP errors with no actionable next steps.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). The 'talk', 'register', 'create_conversation', 'leave_conversation', 'set_status', 'add_friend', 'remove_friend', and 'notifications' tools are write operations, but this is not declared in the schema. LLMs cannot distinguish safe (read-only) from risky (destructive) tools without explicit hints.