Your chatbot went live a few weeks ago. It answers most questions well — but every week you notice the same pattern: a handful of questions it can't quite answer, a persona that feels slightly off, a new product page you keep meaning to add. Fixing each one means logging into the dashboard, hunting through tabs, typing answers by hand, and remembering to come back later to check whether it worked.
Most teams never get around to it. The chatbot stays good enough forever.
There's a better way. BubblaV ships a built-in MCP (Model Context Protocol) server that lets an AI assistant like Claude, ChatGPT, or OpenClaw operate your chatbot for you. You describe what you want in plain language, and your AI does the clicking — diagnosing gaps, writing new knowledge, tuning the bot's personality, setting up human handoff, even building custom tools. The dashboard becomes optional.
From Dashboard to Conversation
Instead of configuring your chatbot through menus and forms, you just talk to your AI. The AI calls BubblaV's MCP tools on your behalf — the same data and actions the dashboard exposes, but driven by natural language.
The old way (manual)
- Open Reports, scan for weak answers
- Copy each question into Knowledge
- Write and paste answers one by one
- Edit persona text in Settings
- Repeat next week
The new way (agentic)
- Ask: “what's my bot struggling with?”
- Ask: “draft answers and add them”
- Ask: “make the tone friendlier”
- Ask: “hand off refunds to a human”
- Done — in one sitting
Which assistants work?
Claude, ChatGPT, and Google Antigravity connect with one-click OAuth (no API key). OpenClaw and any MCP-compatible client connect with an API key. The full setup is in our MCP Server docs.
Step 0: Connect Your Assistant (90 seconds)
You only do this once. Here's the short version — the developer guide has screenshots for every client.
Add the BubblaV connector
In Claude / ChatGPT / Antigravity settings, add a new MCP server with this URL:
https://www.bubblav.com/mcp
Using OpenClaw or another API-key client instead? Generate an MCP API key in Website Settings → API Keys and send it as an X-API-Key header.
Authorize & pick a website
Your assistant opens a BubblaV login page. Sign in and choose which website it may manage. For OAuth clients there are no keys to copy; for API-key clients the key is shown once — save it.
Ask away
Your assistant now sees BubblaV's tools. The recipes below are ready to paste.
Recipe 1 — Diagnose: Find What Your Bot Gets Wrong
Before improving anything, find out where visitors are slipping through. BubblaV already tracks every low-confidence answer and thumbs-down. Hand that data to your AI and ask it to summarize.
Your prompt
“Look at my chatbot's content gaps and most-asked questions from the last 30 days. Group the questions by theme, tell me which topics the bot struggles with most, and rank the top five things I should fix.”
What happens behind the scenes
bubblav_get_content_gaps— questions the bot answered with low confidence or a thumbs-downbubblav_get_most_asked_questions— what visitors actually ask, ranked by frequencybubblav_read_report— the big picture (resolution rate, leads, cost savings) so fixes are prioritized by impact
You get:a short, themed report — “Shipping & returns (8 gaps), Pricing tiers (5 gaps), Integrations (3 gaps)” — instead of scrolling a raw list. The AI does the grouping; you decide what to fix first.
Recipe 2 — Improve: Fill the Gaps Automatically
Now close those gaps. Give your AI the source of truth (a doc, a page, or just your own knowledge) and let it write and add the answers.
Your prompt
“Take the top three content gaps from the shipping theme. Here's our returns policy: [paste text]. Draft a clear Q&A-style answer for each gap, show them to me for approval, then add each one to the knowledge base.”
What happens behind the scenes
bubblav_search_knowledge— checks the gap isn't already covered before writing a duplicatebubblav_add_knowledge— adds each approved answer (chunked & embedded automatically)
Got a PDF or doc instead?
Ask: “Upload this pricing PDF and index it.” Your assistant calls bubblav_upload_knowledge_file (PDF, DOCX, TXT, or Markdown up to 10 MB) and it's extracted and embedded.
Resolved a ticket by hand?
Turn that one-off answer into reusable knowledge: bubblav_sync_ticket_to_knowledge converts a resolved support ticket into a searchable article. (Pro plan and up.)
Worth knowing: the AI drafts and adds the knowledge, but you approve before it goes live. BubblaV doesn't silently rewrite your knowledge base in the background — the improvement loop is something you drive from your assistant, on your schedule.
Recipe 3 — Tune: Change How the Bot Talks
Knowledge is half the battle; tone and behavior are the other half. Your bot's personality lives in its custom instructions(the system prompt), and the widget's words and colors are just as editable — all through MCP.
Your prompt
“Read my bot's current instructions. Then update the persona so it's warmer and more concise, always greets by name, and proactively offers to escalate to a human for anything billing-related. Also change the widget's greeting to ‘Hi there! How can we help you save today?’ and add three starter suggestions.”
What happens behind the scenes
bubblav_get_website_settings→bubblav_update_website_settings— reads then rewritescustom_instructions(the persona, up to 2,000 chars)bubblav_update_widget_appearance— greeting, starter suggestions, colors, position
Because the AI reads the current settings first, it edits in place instead of wiping your config — exactly like a careful human would, just faster.
Recipe 4 — Automate: Handoff & Custom Tools
The dashboard work that takes the longest — wiring up human handoff and building tools — is some of the easiest to delegate. Describe the trigger; your AI configures it.
Your prompt
“Set up human handoff so that whenever a visitor mentions a refund, cancellation, or complaint, the bot routes them to a live agent with the message ‘Let me connect you with a teammate who can sort this out.’ Then create a custom tool called ‘check_order_status' that takes an order number and calls my order-status webhook.”
What happens behind the scenes
bubblav_create_handoff_scenario— the intent-to-human trigger (Pro plan and up)bubblav_create_custom_tool— registers the webhook tool and activates it for your site (Pro plan and up)
Want to go deeper with tools?
Custom tools turn your chatbot from an answerer into a doer — checking orders, updating a CRM, calculating quotes. We wrote a whole guide on letting your AI build them: Create Custom Tools for Your AI Chatbot Without Coding.
Put It in a Loop
These recipes are most powerful chained together into a recurring habit:
// run this whenever you open your assistant
1. Diagnose — “what should I fix this week?”
2. Improve — “draft & add answers for the top gaps”
3. Measure — “how did resolution rate change?”
4. Tune — “adjust tone / handoff based on what you saw”
→ repeat. The bot gets sharper every cycle.
One honest note
Your AI drives each cycle — BubblaV doesn't run an autonomous self-improvement job in the background. That's by design: you stay in control of what gets added and how the persona evolves. The win is that a week's worth of dashboard chores collapses into a single conversation.
Start Delegating
If your chatbot can answer questions, your AI assistant can manage it. Connect Claude, ChatGPT, Antigravity, or OpenClaw and put the recipes above to work this week.
- New to BubblaV? Create a free account and add your site.
- Want the full recipe book with copy-paste prompts? See Use AI Agents to Manage & Improve Your Chatbot in our docs.
- Need the tool-by-tool reference? Browse the MCP Server reference.