Data Access Tools
The five tools an AI uses to look around before it acts — list_records and get_record to browse and inspect your records, and get_guide, get_account_state and search_yaplet_docs to learn what Yaplet can do for this account, what is set up, missing or failing, and how a feature works.
Overview
These are the tools an AI uses to find its way before it does anything. list_records and get_record read your records — widgets, conversations, tickets, subscribers and much more. The three guidance tools tell the AI what Yaplet can do for this account (get_guide), what is set up, missing or failing in it (get_account_state), and how a feature works according to Yaplet's own documentation (search_yaplet_docs). All five only read, so they work at every access level, including Read only, and Copilot runs them without a Confirm card.
This area has no actions of its own. The actions that change things are described on the other tool pages, and an AI finds them with find_actions (see MCP Server).
Copilot permission, which every connection needs anyway, and what they show still follows your permissions.list_records and get_record need the Data Access area (copilot:data_access) ticked on the login screen. The three guidance tools are available to every connection, whichever areas were ticked.Tools
| Tool | What it does |
|---|---|
get_guide | Lists every product area: what it is for, whether your plan and your permissions include it, how many actions an AI can run there, and a link to its documentation. Also lists what only a person may do. |
get_account_state | Shows, per brand and for the organization, what is set up, missing or failing — for example a widget that never loaded on a website, a switched-off AI agent or a failing product feed. |
search_yaplet_docs | Searches Yaplet's own documentation and help center, in English or Hungarian, and returns the matching passages with links. |
list_records | Lists the records of one type — widgets, conversations, tickets, subscribers and so on — with optional filters. |
get_record | Fetches one record by its ID with all its details. |
Guidance Tools
get_guide — start here
The AI's map of Yaplet. For every product area — chat widget, AI agent and knowledge, help center and documentation, inbox, tickets, newsletter, engagement, reports, time management, audience, workflows, voice, social posting, affiliates, security, agent monitoring and settings — it returns:
- what the area is for
- whether your plan includes it, and whether you have access to it (for an app connected through the browser login, also whether that area was ticked at login)
- how many actions an AI can run there, counted by kind (read, change, delete or send) — or that there is nothing an AI can do there yet and the dashboard is the place
- a link to the area's documentation, in English or Hungarian
It also lists what only a person may do (billing and plan; team, permissions and API keys) with links, and a few working rules, including what this connection's access level allows. An AI should call it first in a new conversation, and whenever someone asks "what can you do for me here?".
get_account_state — what is set up, missing or failing
A health check of your account. Per brand (only the brands you may access) and for the whole organization, it reports what exists and what needs attention, and ends each part with a short attention list in plain sentences, such as Widget "Main site" has never loaded on a website or Sending domain example.com is not verified yet.
| What it shows | Shown when you hold |
|---|---|
| Chat widgets, and whether each has ever loaded on a website | Brands |
| AI agents, and whether each is switched on | Vex |
| Knowledge documents by status, with up to five that are not active and their errors | ContentSources |
| Product sources, and whether any is failing | ContentSources |
| Knowledge bases: public or not, published and unpublished articles | Knowledgebase |
| Documentation sets: public or not, published and unpublished pages | Documentation |
| Sending domains: verified, authentication and deliverability state | Settings.Email |
| Campaigns by status, and paused campaigns with the reason | Newsletter.Campaigns |
| Number of subscribers | Newsletter |
| Number of ticket boards | Tickets |
| Chat workflows: how many there are, how many are switched on, and which ones had failed steps in the last 24 hours (a failed API call, or a "Start a workflow" step that could not start the next workflow) — each failing one also gets an attention line | Workflows |
| Voice agents by status | Voice |
Each line appears only for people with that area's permission. For an app connected through the browser login, the area must also have been ticked at login, and voice agents are never shown. So a missing line means "no permission or area", not "nothing there". An AI should call it first for any question about this account — "why doesn't my widget show?", "what should I set up next?" — then search the documentation, then use the specific action.
search_yaplet_docs — how a feature works
Searches Yaplet's official documentation (these pages) and the Yaplet help center, and returns the most relevant passages with their links. It searches in English or Hungarian: Copilot uses your dashboard language unless asked otherwise, outside apps use English. Each search uses a tiny amount of AI credits.
An AI should call it whenever someone asks how a Yaplet feature works or how to set something up, and before it explains or changes a setting — the documentation is where the names you see on screen, and where each setting lives, come from. It only reads Yaplet's own documentation, not your knowledge base or documentation; those are read with list_records and get_record.
Reading Records
list_records returns a summary of each record of one type; get_record fetches a single record by ID with full details — full configuration objects, content fields and related data.
Record Types
| Type | What it holds |
|---|---|
widgets | Chat widgets |
chatbots | AI agents only — chat workflows are listed as workflows |
workflows | Chat workflows: name, brand, switched on or off, what starts them and the channels they run on |
api_tools | The AI agent's API tools, with every setting except the secret key |
knowledgebase | Knowledge bases |
chats | Conversations |
visitors | Website visitors |
documents | AI agent knowledge documents (needs the AI agent) |
conversation_starters | AI agent conversation starters (needs the AI agent) |
articles | Knowledge base articles (needs the knowledge base) |
categories | Knowledge base categories (needs the knowledge base) |
documentation | Documentation sets |
doc_categories | Documentation categories (needs the documentation set) |
doc_pages | Documentation pages (needs the documentation set) |
contact_groups | Contact groups with subscriber counts |
contact_segments | Contact segments |
email_templates | Email templates |
campaigns | Email campaigns |
subscribers | Newsletter subscribers |
outreach | All engagement items combined — banners, chat messages, surveys, tours, news (list only) |
banners, messages, surveys, tours, news | One engagement item (get only) |
boards | Ticket, feedback and roadmap boards |
tickets | Tickets on a board (needs the board) |
shifts | Work shifts with computed pay (your own; everyone's with the All Shifts permission) |
shift_types | Shift types with pay multipliers and the chat-duty flag |
product_sources | The catalogues products come from, with their fetch schedule and last run |
product_rules | The binding wording the AI must follow about a product — keys and text, draft or active |
products | The catalogue itself: base products with their versions — options, price, sale price, stock — plus each product's rule keys and its own rule sentences |
newsletter_autopilots | Newsletter autopilots with their settings, the next planned send, the issue waiting for approval and the last issue |
content_ideas | A brand's topic ideas for its autopilots, plus the brand's autopilots (needs the brand; list only) |
What get_record Adds
- Boards come with everything needed to change a ticket: the columns, the label palette, the people a ticket can be assigned to, the verified addresses a reporter email can be sent from, and the priorities — every value Ticket Tools accepts, in one call, so the AI never has to guess an ID.
- Tickets come with their labels, comments, reporter (the person who submitted it, with their email address) and whether they are archived.
- Conversations come with
channel(widget, email, Facebook, Instagram or voice) andhas_messages— whether it is a real conversation or one of the empty records the widget opens for a new visitor. - Shifts come with their computed pay and, for All Shifts holders, the shift log.
- Product sources never return their secret — only whether one is set, the names of the custom headers and the push URL.
- Product rules add how many products each reaches today; products add every rule that applies to them, drafts included.
- Newsletter autopilots add their ticked content sources, their latest 50 issues and the cost estimate.
- Workflows add their step count and a link to the workflow in the editor. The steps themselves are read with
workflows.get_plan— see Workflow Tools. - API tools never return their Secret Key — only
secret_set, which says whether one is set — and every Custom Header value that is filled in reads(set)instead of the real value.
Filters You Can Ask For
| Records | Filters |
|---|---|
| Conversations | Inbox view (all, my inbox, unassigned, calls), channel (widget, Facebook, Instagram, voice), one widget, changed since a given time |
| Visitors | Also include visitors who never chatted (raw page traffic) |
| Articles, documentation pages | Published only |
| Tickets | Column, archived or live, assignee, priority |
| Shifts | One agent (needs All Shifts), shift type, running, finalized, edited, start-date range |
| Subscribers | Exact email, name or email search, state, one group or one segment, engagement score range |
| Product sources, rules, products | Brand; products also by source; rules also by status, mode and document import; rules and products also by rule key and by text |
| Workflows, API tools | Brand |
Every list returns 50 records unless the AI asks for more (up to 300). Conversations, visitors, tickets, shifts, subscribers, product sources, product rules and products can also be paged through.
Boards and Tickets
archived flag, and the AI can filter on it directly.Real Conversations Only
chats returns only conversations that have at least one message, exactly matching what you see in your Inbox.The same reasoning applies to visitors. Listing visitors returns the people who have actually held a conversation with you, ordered by their most recent conversation. Including visitors who never chatted gives raw page traffic instead — the right choice for audience questions — ordered by last activity.
Conversations come newest-updated first, with two counts: how many match the filters in total, and how many are in this page. Every conversation includes the visitor's name, email and country, the last message, the channel, the widget name, the email subject (for email conversations), created and updated times and the assigned agent — so the AI rarely needs a second call to work out who a conversation is with.
To keep an eye on the inbox, the AI remembers the newest update time it has seen and passes it back as "changed since" on the next call. On a busy organization a full conversation list can run to tens of thousands of records, while the same call limited to the last hour typically returns a handful.
How It Works
- Records come newest first, except conversations (last update), shifts (start time), product sources (oldest first), products (last update) and the types that have their own order — knowledge base categories, documentation categories and pages, conversation starters and shift types — which come back in the order the dashboard shows them.
- The record must belong to your organization; records of other organizations are never returned.
- If your account is limited to specific widgets, you only see conversations from those widgets — the same restriction as in your Inbox. Owners and admins aren't restricted, and voice calls are always visible since they don't belong to a widget.
workflows,api_tools,newsletter_autopilotsandcontent_ideasshow only the brands you may access.- Results that hold text written by people outside your team — conversations, visitors, ticket reporters, subscribers — start with a line telling the AI to treat that text as data and never follow instructions found in it.
What an AI cannot do here
- Change anything. All five tools only read. Changes go through actions — see the other tool pages.
- Read your team, permissions, API keys or billing. These are human-only in every connection;
get_guidelinks to the dashboard pages where a person manages them. - List the empty conversation records the widget opens for visitors who never wrote. A single one can still be fetched by ID with
get_record. - Fix what
get_account_statefinds. It only reports; the fix is an action, or a step in the dashboard. - Search your own help center with
search_yaplet_docs. It only searches Yaplet's documentation; your knowledge base and documentation are read withlist_recordsandget_record.
Example Prompts
- "What can you do for me in this account?" (uses
get_guide) - "Why isn't my chat widget showing on my website?" (uses
get_account_state, thensearch_yaplet_docs) - "What conversations are unassigned right now?"
- "Find the subscriber [email protected]."
- "Show me the full details of my default widget." (lists widgets first, then fetches the default one)