Corebee Mcp Skills — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited Corebee Mcp Skills (Plugin) and scored it 96/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 1 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 1 flagged
A bulleted imperative like {match} tells the agent to never reveal, disclose, or mention something to the user. Used adversarially it can instruct the agent to hide its tool calls or lie about what it did — stripping the transparency a user relies on to trust the agent.
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
Corebee is an AI-powered customer support platform with conversations, knowledge base, contact management, analytics, and multi-channel support. This plugin gives Claude full operational control over a Corebee workspace through 9 skills, 4 specialized agents, 3 slash commands, and a hook system for guardrails -- making it the most comprehensive support operations plugin available for Claude Code and Claude.ai.
/plugin marketplace add jonny-1812/corebee-mcp-skills
/plugin install corebee@corebeeThen run /mcp and approve the Corebee OAuth login. No keys to paste — OAuth is discovered automatically. You get the hosted MCP server plus 9 skills, 4 agents, and 3 commands.
claude mcp add --transport http corebee https://corebee.ai/api/v1/mcp/streamAdd Corebee as a remote MCP connector:
https://corebee.ai/api/v1/mcp/streamTell your agent: "Set up Corebee AI customer support for my product — start the free trial." It signs you up via https://corebee.ai/api/auth/agent-signup, you approve one email (with a pairing phrase), and it hands back your widget snippet. 14-day trial, no card. Full agent flow: <https://corebee.ai/for-ai-agents> · <https://corebee.ai/llms.txt>
| Component | Count | Description |
|---|---|---|
| Skills | 9 | Guided workflows for common support operations |
| Agents | 4 | Specialized sub-agents for delegated analysis and action |
| Commands | 3 | Slash commands for quick, frequent tasks |
| Hooks | 5 | Guardrail scripts that validate, warn, and log tool usage |
| Settings | 6 | Configurable defaults shared across all skills |
Skills are structured workflows that guide Claude through multi-step support operations using the MCP tools.
| Skill | Description | Example Triggers |
|---|---|---|
| inbox-triage | Prioritize open conversations by urgency, suggest assignments, auto-assign with confirmation | "Triage my inbox", "What needs attention?" |
| weekly-report | Generate a full support report with KPIs, team rankings, channel breakdown, and trends | "Weekly report", "How did we do last week?" |
| contact-onboard | Add contacts with duplicate detection, bulk import, updates, and contextual notes | "Add John Smith [email protected]", "Onboard these 5 contacts" |
| knowledge-manager | List sources, add URLs, search content, audit coverage gaps, view indexing stats | "What's in my KB?", "Add https://docs.acme.com" |
| widget-customizer | Configure chat widget colors, bot name, position, welcome message, and pre-chat form | "Make the widget blue", "Set bot name to Bee" |
| escalation-manager | Detect urgent conversations, assign to the best agent, log escalation notes, draft responses | "Escalate conversation 123", "Scan for anything urgent" |
| team-workload | Analyze agent capacity, detect overload, rebalance conversation assignments | "Show team workload", "Rebalance the queue" |
| metrics-compare | Compare support metrics between two periods with percentage deltas and trend analysis | "Compare this week to last week", "Is response time improving?" |
| customer-360 | Build a unified customer profile from contacts, conversations, notes, and KB articles | "Tell me about [email protected]", "Prepare me for a call with Acme" |
Agents are specialized sub-agents that Claude delegates to for focused tasks. Each agent has a defined toolset and methodology.
| Agent | Role | Tools | Delegated When |
|---|---|---|---|
| support-analyst | Read-only analytics with a 5-step analysis methodology (collect, compare, correlate, conclude, recommend) | 5 | User asks for deep metric analysis, trend explanations, or data-driven recommendations |
| conversation-helper | Conversation lifecycle management with tone detection, KB-backed replies, and batch processing | 9 | User wants to draft replies, review threads, close tickets, or manage assignments |
| kb-optimizer | Knowledge base gap analysis by cross-referencing conversations with KB content | 7 | User wants to find what the AI cannot answer, identify outdated content, or improve KB coverage |
| sla-monitor | SLA compliance monitoring with per-agent compliance rates and corrective action suggestions | 7 | User wants to check for SLA breaches, identify at-risk conversations, or monitor response times |
Commands are lightweight slash-command shortcuts for frequent operations.
| Command | Usage | What It Does |
|---|---|---|
/corebee:triage | Run directly | Quick inbox triage -- classifies conversations as urgent, stale, or active and suggests assignments |
/corebee:report | Run directly | Generates a weekly support report with overview KPIs, team performance table, channel breakdown, and trends |
/corebee:kb-status | Run directly | Audits knowledge base health -- counts sources by status, flags errors and empty sources, recommends fixes |
Hooks are shell scripts that run automatically before or after specific tool calls, adding guardrails to support operations.
| Hook | Trigger | Behavior |
|---|---|---|
| validate-reply.sh | Before reply_to_conversation | Blocks empty replies, replies under 10 characters, and replies over 5,000 characters |
| warn-destructive.sh | Before delete_* or close_* | Displays a warning describing the destructive action about to be taken |
| remind-resolution-note.sh | After close_conversation | Reminds the user to add a resolution note to the contact record |
| confirm-assignment.sh | After assign_conversation | Confirms the assignment and suggests checking team workload for balanced distribution |
| log-agent-start.sh | On sub-agent delegation | Logs which agent is being delegated to for visibility |
All skills share a common settings file (settings.json) with configurable defaults.
| Setting | Type | Default | Description |
|---|---|---|---|
default_period | string | this_week | Default time period for metrics and reports. Options: today, this_week, this_month, last_7_days, last_30_days |
sla_threshold_hours | number | 4 | Hours before an unanswered conversation is flagged as SLA-breaching (range: 1-72) |
auto_assign | boolean | false | Whether skills execute assignments without confirmation |
report_format | string | markdown | Output format for reports. Options: markdown, plain_text |
show_ids | boolean | false | Display internal UUIDs alongside human-readable names |
timezone | string | UTC | IANA timezone for all timestamps and date calculations |
Server URL: https://corebee.ai/api/v1/mcp/stream (Streamable HTTP, OAuth 2.1 auto-discovered)
The Corebee MCP server exposes 28 tools across 5 categories:
| Category | Tools | Description |
|---|---|---|
| Conversations | 5 | List, view, reply, assign, and close support conversations |
| Contacts | 6 | Manage customer contacts -- search, create, update, delete, and add notes |
| Knowledge Base | 6 | Manage AI training data -- add URLs, search content, track indexing |
| Metrics | 5 | Dashboard KPIs, agent performance, channel analytics, trends, reports |
| Settings | 6 | Widget customization, team members, organization settings |
All 28 tools are referenced by at least one skill or agent in this plugin.
Inbox and triage:
Reporting and analytics:
Conversations:
Contacts and customers:
Knowledge base:
Team and workload:
Widget:
The MCP server supports two authentication methods:
X-API-Key header -- for programmatic access and custom integrationsMIT
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.