Tools & toolsets
About 4 minutesUpdated August 2026
Every Qalyb MCP tool, grouped into toolsets you can enable per connection, and the guardrails they all follow.
| Toolset | What it does | Tools |
|---|---|---|
| core | Identity, members, plan & usage. | whoami, list_members, get_usage_summary |
| agents | Build and manage agents. | list_agents, get_agent, create_agent, update_agent, list/restore prompt revisions, delete_agent, get/set escalation policy, set success criteria, set extraction schema, list templates, personas, voices |
| conversations | Calls, chats, transcripts, recordings, exports. | list_calls, search_transcripts, get_call, get_call_recording, get_call_execution_log, start_outbound_call, delete_calls, exports |
| tickets | The human-handoff inbox. | list_tickets, get_ticket, get_ticket_counts, get_escalation_analytics, claim_ticket, assign_ticket, reply_to_ticket, resolve_ticket, reopen_ticket, add_ticket_note, set_ticket_status, set_ticket_priority, create_ticket |
| analytics | Timezone-aware performance metrics. | get_analytics_overview, get_channel_analytics, get_call_analytics, get_agent_analytics, get_campaign_analytics, get_chat_analytics |
| campaigns | Outbound voice / WhatsApp at scale. | list_campaigns, get_campaign, create_campaign, update_campaign, add_campaign_contacts, cancel_campaign, delete_campaign, list_campaign_calls, retry_campaign_call |
| knowledge | RAG knowledge bases and documents. | list_knowledge_bases, create_knowledge_base, update_knowledge_base, list/add documents, get/update document content, query_knowledge_base, delete document / base |
| components | The UI components your agents render in chat. | list_components, get_component, create_component, update_component, clone_component, archive_component |
| tools | The API calls your agents make. Webhook tools only — integration tools are created by the Shopify, Salla and Composio connect flows, together with the credentials that make them work. Header values are never returned, only header names. | list_tools, get_tool, create_tool, update_tool, toggle_tool, test_tool, delete_tool |
| channels | A read-only view of where your agents are actually live: widgets, messaging channels, phone numbers and SIP trunks. SIP credentials are reported as present-or-absent, never as values. | list_channels, get_channel, list_phone_numbers, list_sip_trunks, list_widget_tokens |
| docs | Search the public Qalyb documentation. | search_documentation |
Searching the documentation
search_documentation searches these guides — the public documentation at qalyb.ai/docs — and returns the matching passages together with the URL of the page each came from, so a client can cite and link its answer. Ask it how something works: "how do I connect WhatsApp", "what does this setting do", "which channels are supported".
It is the one tool that does not read your workspace. The corpus is the published documentation, identical for every organization, so the tool needs no permission and takes no knowledge-base argument — there is no way to point it at anything else. To search your own content, use query_knowledge_base in the knowledge toolset instead.
Guardrails
- Destructive actions confirm first. Deleting an agent or cancelling a campaign without
confirm: truereturns a preview (including deletion impact) instead of executing. - Prompt edits are reversible.
update_agentreplaces prompt text by default — passprompt_mode: "append"to add to it instead. A replace that would cut the prompt to under a third of its length needsconfirm_replace: true, and every change from any surface is snapshotted, sorestore_agent_prompt_revisioncan put the old prompt back. - Shared templates stay shared. The global component gallery is readable and clonable from your workspace but never editable through it —
clone_componentgives you your own copy to change. - Deleting needs an admin. Agents, knowledge bases, knowledge documents and components can only be deleted by a workspace admin, exactly as on the dashboard — a connection signed in as a member never sees those tools. API keys act as workspace admin, so they do.
- Names work everywhere. Tools accept agent, member, and knowledge-base names — not just ids — and disambiguate when several match.
- Workspace isolation. Every tool is scoped to the authenticated workspace; other organizations' data is invisible. The one exception is
search_documentation, which reads the public documentation and no workspace data at all. - Plan limits speak up. Hitting a plan gate returns a clear upgrade message rather than a cryptic error.
Building an integration in code instead? The same operations are on the REST API.
Was this guide helpful?